diff --git a/CLAUDE.md b/CLAUDE.md index 302cdeab99..f2bd317322 100644 --- a/CLAUDE.md +++ b/CLAUDE.md @@ -1 +1,36 @@ -IMPORTANT: Ensure you’ve thoroughly reviewed the [AGENTS.md](AGENTS.md) file before beginning any work. +IMPORTANT: Ensure you've thoroughly reviewed the [AGENTS.md](AGENTS.md) file before beginning any work. + +# AI Policy +- AI is assistive only; AI-generated PRs are restricted per AGENTS.md +- Contributor reviews and writes code themselves + +# Code Style & Conventions +- snake_case naming; optimize for longest common prefix +- 4 spaces indentation, brackets on same line +- `void * ptr`, `int & a` (space around pointer/reference) +- Avoid templates and fancy STL +- Use sized integer types (`int32_t`) in public API +- See [CONTRIBUTING.md](CONTRIBUTING.md) for full guidelines, naming, and PR process + +# ggml Tensor Conventions +- Data stored in row-major order +- Dimension 0 = columns, dimension 1 = rows, dimension 2 = matrices +- **Matrix multiply is unconventional**: `C = ggml_mul_mat(ctx, A, B)` means `C^T = A * B^T` + +# Quantization +- See [docs/quantization/](docs/quantization/) for comprehensive documentation +- See [docs/quantization/09-adding-new-types.md](docs/quantization/09-adding-new-types.md) for adding new types + +## Key Files +- `ggml/include/ggml.h`: type enums (`ggml_type`) +- `ggml/src/ggml-common.h`: block structures +- `ggml/src/ggml-quants.c`: reference quantize/dequantize implementations +- `tools/quantize/quantize.cpp`: CLI tool +- `src/llama-quant.cpp`: core quantization engine + +## Quantization Families +- **Q**: simple uniform quantization +- **K**: super-block quantization (multiple sub-blocks per super-block) +- **IQ**: importance-weighted quantization +- **T**: ternary quantization +- **MXFP**: Microsoft floating-point quantization diff --git a/src/llama-quant.cpp b/src/llama-quant.cpp index 65e5452e2f..e5890936c4 100644 --- a/src/llama-quant.cpp +++ b/src/llama-quant.cpp @@ -13,22 +13,6 @@ #include #include -// tensor categorization - used to avoid repeated string matching in quantization logic. -// this is different from LLM_TN - we want broad categories, not specific tensor names per arch. -enum class tensor_category { - TOKEN_EMBD, - ATTENTION_Q, - ATTENTION_V, - ATTENTION_K, - ATTENTION_QKV, - ATTENTION_KV_B, - ATTENTION_OUTPUT, - FFN_UP, - FFN_GATE, - FFN_DOWN, - OUTPUT, - OTHER -}; static void zeros(std::ofstream & file, size_t n) { char zero = 0; @@ -150,15 +134,6 @@ static bool category_is_attn_v(tensor_category cat) { cat == tensor_category::ATTENTION_KV_B; } -// per-tensor metadata, computed in the preliminary loop and used in the main loop -struct tensor_metadata { - ggml_type target_type; - tensor_category category; - std::string remapped_imatrix_name; - bool allows_quantization; - bool requires_imatrix; -}; - // // dequantization // @@ -355,6 +330,7 @@ static ggml_type tensor_type_fallback(quantize_state_impl & qs, const ggml_tenso return return_type; } +// internal standard logic for selecting the target tensor type based on tensor category, ftype, and model arch static ggml_type llama_tensor_get_type_impl(quantize_state_impl & qs, ggml_type new_type, const ggml_tensor * tensor, llama_ftype ftype, tensor_category category) { const std::string name = ggml_get_name(tensor); @@ -604,15 +580,8 @@ static ggml_type llama_tensor_get_type_impl(quantize_state_impl & qs, ggml_type return new_type; } -// public API: compute category from tensor name and delegate to _impl -ggml_type llama_tensor_get_type(quantize_state_impl & qs, ggml_type new_type, const ggml_tensor * tensor, llama_ftype ftype) { - const std::string name = ggml_get_name(tensor); - tensor_category category = tensor_get_category(name); - return llama_tensor_get_type_impl(qs, new_type, tensor, ftype, category); -} - -// outer wrapper: determine the ggml_type that this tensor should be quantized to (used internally by llama_model_quantize_impl) -static ggml_type llama_tensor_get_type_internal(quantize_state_impl & qs, const llama_model_quantize_params * params, const ggml_tensor * tensor, ggml_type default_type, const tensor_metadata & tm) { +// outer wrapper: determine the ggml_type that this tensor should be quantized to +ggml_type llama_tensor_get_type(quantize_state_impl & qs, const llama_model_quantize_params * params, const ggml_tensor * tensor, ggml_type default_type, const tensor_metadata & tm) { if (!tensor_allows_quantization(params, qs.model.arch, tensor)) { return tensor->type; } @@ -845,15 +814,16 @@ const char * llama_ftype_to_name(llama_ftype ftype) { return nullptr; } -void init_quantize_state_counters(quantize_state_impl & qs, const std::vector & tensor_names) { - for (const auto & name : tensor_names) { - tensor_category cat = tensor_get_category(name); +void init_quantize_state_counters(quantize_state_impl & qs, std::vector & metadata) { + for (auto & tm : metadata) { + tensor_category cat = tensor_get_category(tm.name); + tm.category = cat; if (category_is_attn_v(cat)) { ++qs.n_attention_wv; } - if (tensor_name_match_output_weight(name.c_str())) { + if (cat == tensor_category::OUTPUT) { qs.has_tied_embeddings = false; } } @@ -996,15 +966,15 @@ static void llama_model_quantize_impl(const std::string & fname_inp, const std:: }); } - { - std::vector tensor_names; - tensor_names.reserve(tensors.size()); - for (const auto * it : tensors) { - tensor_names.emplace_back(ggml_get_name(it->tensor)); - } - init_quantize_state_counters(qs, tensor_names); + // compute tensor metadata once and cache it + std::vector metadata(tensors.size()); + for (size_t i = 0; i < tensors.size(); ++i) { + metadata[i].name = ggml_get_name(tensors[i]->tensor); } + // initialize quantization state counters and metadata categories + init_quantize_state_counters(qs, metadata); + int idx = 0; uint16_t n_split = 1; @@ -1017,25 +987,6 @@ static void llama_model_quantize_impl(const std::string & fname_inp, const std:: std::vector ctx_outs(n_split); ctx_outs[0] = std::move(ctx_out); - // compute tensor metadata once and cache it - std::vector metadata(tensors.size()); - - // initialize quantization state before preliminary loop (counters for use_more_bits) - { - for (size_t i = 0; i < tensors.size(); ++i) { - const auto cat = tensor_get_category(tensors[i]->tensor->name); - if (category_is_attn_v(cat)) { - ++qs.n_attention_wv; - } - if (cat == tensor_category::OUTPUT) { - qs.has_tied_embeddings = false; - } - metadata[i].category = cat; // save and re-use the category while we're at it - } - // these also need to be set to n_layer by default - qs.n_ffn_down = qs.n_ffn_gate = qs.n_ffn_up = (int)qs.model.hparams.n_layer; - } - // flag for --dry-run bool will_require_imatrix = false; @@ -1059,7 +1010,7 @@ static void llama_model_quantize_impl(const std::string & fname_inp, const std:: metadata[i].allows_quantization = tensor_allows_quantization(params, model.arch, tensor); if (metadata[i].allows_quantization) { - metadata[i].target_type = llama_tensor_get_type_internal(qs, params, tensor, default_type, metadata[i]); + metadata[i].target_type = llama_tensor_get_type(qs, params, tensor, default_type, metadata[i]); } else { metadata[i].target_type = tensor->type; } diff --git a/src/llama-quant.h b/src/llama-quant.h index f3c1290c58..d8fe643557 100644 --- a/src/llama-quant.h +++ b/src/llama-quant.h @@ -12,6 +12,33 @@ struct llama_model; +// tensor categorization - used to avoid repeated string matching in quantization logic. +// this is different from LLM_TN - we want broad categories, not specific tensor names per arch. +enum class tensor_category { + TOKEN_EMBD, + ATTENTION_Q, + ATTENTION_V, + ATTENTION_K, + ATTENTION_QKV, + ATTENTION_KV_B, + ATTENTION_OUTPUT, + FFN_UP, + FFN_GATE, + FFN_DOWN, + OUTPUT, + OTHER +}; + +// per-tensor metadata, computed in the preliminary loop and used in the main loop +struct tensor_metadata { + std::string name; + ggml_type target_type; + tensor_category category; + std::string remapped_imatrix_name; + bool allows_quantization; + bool requires_imatrix; +}; + // result of parsing --tensor-type option // (changes to this struct must be reflected in tools/quantize/quantize.cpp) struct tensor_type_option { @@ -56,7 +83,7 @@ struct quantize_state_impl { } }; -ggml_type llama_tensor_get_type(quantize_state_impl & qs, ggml_type new_type, const ggml_tensor * tensor, llama_ftype ftype); +ggml_type llama_tensor_get_type(quantize_state_impl & qs, const llama_model_quantize_params * params, const ggml_tensor * tensor, ggml_type default_type, const tensor_metadata & tm); ggml_type llama_ftype_get_default_type(llama_ftype ftype); // Ftype name <-> enum conversions. @@ -64,9 +91,9 @@ ggml_type llama_ftype_get_default_type(llama_ftype ftype); llama_ftype llama_ftype_from_name(const char * name); const char * llama_ftype_to_name(llama_ftype ftype); -// Initialize quantize_state_impl counters by scanning tensor names. -// tensor_names: all quantizable weight tensor names in the model. -void init_quantize_state_counters(quantize_state_impl & qs, const std::vector & tensor_names); +// Initialize quantize_state_impl counters and populate tensor_metadata categories. +// metadata: vector with name fields already set, will have category field populated. +void init_quantize_state_counters(quantize_state_impl & qs, std::vector & metadata); // Returns true if this tensor should be quantized (based on name, dims, params). bool tensor_allows_quantization(const llama_model_quantize_params * params, llm_arch arch, const ggml_tensor * tensor); diff --git a/tests/gguf-model-data.cpp b/tests/gguf-model-data.cpp index 97de083568..61801256c6 100644 --- a/tests/gguf-model-data.cpp +++ b/tests/gguf-model-data.cpp @@ -515,7 +515,8 @@ static std::string detect_gguf_filename(const std::string & repo, const std::str static std::optional fetch_and_parse( const std::string & repo, const std::string & filename, - const std::string & cache_path) { + const std::string & cache_path, + bool verbose) { std::string url = "https://huggingface.co/" + repo + "/resolve/main/" + filename; // Progressive download inspired by RangeView.fetchChunk() @@ -524,7 +525,9 @@ static std::optional fetch_and_parse( const size_t max_chunk = 64 * 1024 * 1024; while (chunk_size <= max_chunk) { - fprintf(stderr, "gguf_fetch: downloading %zu bytes from %s\n", chunk_size, filename.c_str()); + if (verbose) { + fprintf(stderr, "gguf_fetch: downloading %zu bytes from %s\n", chunk_size, filename.c_str()); + } char range_buf[64]; snprintf(range_buf, sizeof(range_buf), "bytes=0-%zu", chunk_size - 1); @@ -565,7 +568,8 @@ static std::optional fetch_or_cached( const std::string & repo, const std::string & filename, const std::string & cdir, - const std::string & repo_part) { + const std::string & repo_part, + bool verbose) { std::string fname_part = sanitize_for_path(filename); std::string cache_path = cdir + "/" + repo_part + "--" + fname_part + ".partial"; @@ -574,20 +578,23 @@ static std::optional fetch_or_cached( if (std::filesystem::exists(cache_path) && read_file(cache_path, cached)) { auto result = gguf_parse_meta(cached); if (result.has_value()) { - fprintf(stderr, "gguf_fetch: loaded from cache: %s\n", cache_path.c_str()); + if (verbose) { + fprintf(stderr, "gguf_fetch: loaded from cache: %s\n", cache_path.c_str()); + } return result; } } } fs_create_directory_with_parents(cdir); - return fetch_and_parse(repo, filename, cache_path); + return fetch_and_parse(repo, filename, cache_path, verbose); } std::optional gguf_fetch_model_meta( const std::string & repo, const std::string & quant, - const std::string & cache_dir) { + const std::string & cache_dir, + bool verbose) { std::string cdir = cache_dir.empty() ? get_default_cache_dir() : cache_dir; std::string repo_part = sanitize_for_path(repo); @@ -597,7 +604,7 @@ std::optional gguf_fetch_model_meta( return std::nullopt; } - auto model_opt = fetch_or_cached(repo, filename, cdir, repo_part); + auto model_opt = fetch_or_cached(repo, filename, cdir, repo_part, verbose); if (!model_opt.has_value()) { fprintf(stderr, "gguf_fetch: failed to fetch %s\n", filename.c_str()); return std::nullopt; @@ -612,8 +619,10 @@ std::optional gguf_fetch_model_meta( return std::nullopt; } - fprintf(stderr, "gguf_fetch: split model with %u shards, fetching remaining %u...\n", - model.n_split, model.n_split - 1); + if (verbose) { + fprintf(stderr, "gguf_fetch: split model with %u shards, fetching remaining %u...\n", + model.n_split, model.n_split - 1); + } for (int i = 2; i <= model.n_split; i++) { char num_buf[6], total_buf[6]; @@ -621,7 +630,7 @@ std::optional gguf_fetch_model_meta( snprintf(total_buf, sizeof(total_buf), "%05d", (int)model.n_split); std::string shard_name = split_prefix + "-" + num_buf + "-of-" + total_buf + ".gguf"; - auto shard = fetch_or_cached(repo, shard_name, cdir, repo_part); + auto shard = fetch_or_cached(repo, shard_name, cdir, repo_part, verbose); if (!shard.has_value()) { fprintf(stderr, "gguf_fetch: failed to fetch shard %d: %s\n", i, shard_name.c_str()); return std::nullopt; diff --git a/tests/gguf-model-data.h b/tests/gguf-model-data.h index ed433791ad..c49c7c6fbe 100644 --- a/tests/gguf-model-data.h +++ b/tests/gguf-model-data.h @@ -39,4 +39,5 @@ struct gguf_remote_model { std::optional gguf_fetch_model_meta( const std::string & repo, const std::string & quant = "Q8_0", - const std::string & cache_dir = ""); // empty = default + const std::string & cache_dir = "", // empty = default + bool verbose = true); diff --git a/tests/snapshots/deepseek-v3.1.schema b/tests/snapshots/deepseek-v3.1.schema index a5bce29b27..0e9dfd6ed4 100644 --- a/tests/snapshots/deepseek-v3.1.schema +++ b/tests/snapshots/deepseek-v3.1.schema @@ -2,10 +2,8 @@ # n_embd=7168, n_ff=18432, n_vocab=129280, n_layer=61, n_head=128, n_head_kv=1, n_expert=256 [F32] f32 -output.weight q6_K [F16] f16 -output.weight q6_K [Q4_0] q4_0 output.weight q6_K @@ -23,1098 +21,1952 @@ output.weight q6_K [Q2_K] q2_K output.weight q6_K +blk.0.attn_k_b.weight q4_0 blk.0.attn_output.weight q3_K blk.0.ffn_down.weight q3_K +blk.1.attn_k_b.weight q4_0 blk.1.attn_output.weight q3_K blk.1.ffn_down.weight q3_K +blk.2.attn_k_b.weight q4_0 blk.2.attn_output.weight q3_K blk.2.ffn_down.weight q3_K +blk.3.attn_k_b.weight q4_0 blk.3.attn_output.weight q3_K blk.3.ffn_down_exps.weight q3_K blk.3.ffn_down_shexp.weight q3_K +blk.4.attn_k_b.weight q4_0 blk.4.attn_output.weight q3_K blk.4.ffn_down_exps.weight q3_K blk.4.ffn_down_shexp.weight q3_K +blk.5.attn_k_b.weight q4_0 blk.5.attn_output.weight q3_K blk.5.ffn_down_exps.weight q3_K blk.5.ffn_down_shexp.weight q3_K +blk.6.attn_k_b.weight q4_0 blk.6.attn_output.weight q3_K blk.6.ffn_down_exps.weight q3_K blk.6.ffn_down_shexp.weight q3_K +blk.7.attn_k_b.weight q4_0 blk.7.attn_output.weight q3_K blk.7.ffn_down_exps.weight q3_K blk.7.ffn_down_shexp.weight q3_K +blk.8.attn_k_b.weight q4_0 blk.8.attn_output.weight q3_K blk.8.ffn_down_exps.weight q3_K blk.8.ffn_down_shexp.weight q3_K +blk.9.attn_k_b.weight q4_0 blk.9.attn_output.weight q3_K blk.9.ffn_down_exps.weight q3_K blk.9.ffn_down_shexp.weight q3_K +blk.10.attn_k_b.weight q4_0 blk.10.attn_output.weight q3_K blk.10.ffn_down_exps.weight q3_K blk.10.ffn_down_shexp.weight q3_K +blk.11.attn_k_b.weight q4_0 blk.11.attn_output.weight q3_K blk.11.ffn_down_exps.weight q3_K blk.11.ffn_down_shexp.weight q3_K +blk.12.attn_k_b.weight q4_0 blk.12.attn_output.weight q3_K blk.12.ffn_down_exps.weight q3_K blk.12.ffn_down_shexp.weight q3_K +blk.13.attn_k_b.weight q4_0 blk.13.attn_output.weight q3_K blk.13.ffn_down_exps.weight q3_K blk.13.ffn_down_shexp.weight q3_K +blk.14.attn_k_b.weight q4_0 blk.14.attn_output.weight q3_K blk.14.ffn_down_exps.weight q3_K blk.14.ffn_down_shexp.weight q3_K +blk.15.attn_k_b.weight q4_0 blk.15.attn_output.weight q3_K blk.15.ffn_down_exps.weight q3_K blk.15.ffn_down_shexp.weight q3_K +blk.16.attn_k_b.weight q4_0 blk.16.attn_output.weight q3_K blk.16.ffn_down_exps.weight q3_K blk.16.ffn_down_shexp.weight q3_K +blk.17.attn_k_b.weight q4_0 blk.17.attn_output.weight q3_K blk.17.ffn_down_exps.weight q3_K blk.17.ffn_down_shexp.weight q3_K +blk.18.attn_k_b.weight q4_0 blk.18.attn_output.weight q3_K blk.18.ffn_down_exps.weight q3_K blk.18.ffn_down_shexp.weight q3_K +blk.19.attn_k_b.weight q4_0 blk.19.attn_output.weight q3_K blk.19.ffn_down_exps.weight q3_K blk.19.ffn_down_shexp.weight q3_K +blk.20.attn_k_b.weight q4_0 blk.20.attn_output.weight q3_K blk.20.ffn_down_exps.weight q3_K blk.20.ffn_down_shexp.weight q3_K +blk.21.attn_k_b.weight q4_0 blk.21.attn_output.weight q3_K blk.21.ffn_down_exps.weight q3_K blk.21.ffn_down_shexp.weight q3_K +blk.22.attn_k_b.weight q4_0 blk.22.attn_output.weight q3_K blk.22.ffn_down_exps.weight q3_K blk.22.ffn_down_shexp.weight q3_K +blk.23.attn_k_b.weight q4_0 blk.23.attn_output.weight q3_K blk.23.ffn_down_exps.weight q3_K blk.23.ffn_down_shexp.weight q3_K +blk.24.attn_k_b.weight q4_0 blk.24.attn_output.weight q3_K blk.24.ffn_down_exps.weight q3_K blk.24.ffn_down_shexp.weight q3_K +blk.25.attn_k_b.weight q4_0 blk.25.attn_output.weight q3_K blk.25.ffn_down_exps.weight q3_K blk.25.ffn_down_shexp.weight q3_K +blk.26.attn_k_b.weight q4_0 blk.26.attn_output.weight q3_K blk.26.ffn_down_exps.weight q3_K blk.26.ffn_down_shexp.weight q3_K +blk.27.attn_k_b.weight q4_0 blk.27.attn_output.weight q3_K blk.27.ffn_down_exps.weight q3_K blk.27.ffn_down_shexp.weight q3_K +blk.28.attn_k_b.weight q4_0 blk.28.attn_output.weight q3_K blk.28.ffn_down_exps.weight q3_K blk.28.ffn_down_shexp.weight q3_K +blk.29.attn_k_b.weight q4_0 blk.29.attn_output.weight q3_K blk.29.ffn_down_exps.weight q3_K blk.29.ffn_down_shexp.weight q3_K +blk.30.attn_k_b.weight q4_0 blk.30.attn_output.weight q3_K blk.30.ffn_down_exps.weight q3_K blk.30.ffn_down_shexp.weight q3_K +blk.31.attn_k_b.weight q4_0 blk.31.attn_output.weight q3_K blk.31.ffn_down_exps.weight q3_K blk.31.ffn_down_shexp.weight q3_K +blk.32.attn_k_b.weight q4_0 blk.32.attn_output.weight q3_K blk.32.ffn_down_exps.weight q3_K blk.32.ffn_down_shexp.weight q3_K +blk.33.attn_k_b.weight q4_0 blk.33.attn_output.weight q3_K blk.33.ffn_down_exps.weight q3_K blk.33.ffn_down_shexp.weight q3_K +blk.34.attn_k_b.weight q4_0 blk.34.attn_output.weight q3_K blk.34.ffn_down_exps.weight q3_K blk.34.ffn_down_shexp.weight q3_K +blk.35.attn_k_b.weight q4_0 blk.35.attn_output.weight q3_K blk.35.ffn_down_exps.weight q3_K blk.35.ffn_down_shexp.weight q3_K +blk.36.attn_k_b.weight q4_0 blk.36.attn_output.weight q3_K blk.36.ffn_down_exps.weight q3_K blk.36.ffn_down_shexp.weight q3_K +blk.37.attn_k_b.weight q4_0 blk.37.attn_output.weight q3_K blk.37.ffn_down_exps.weight q3_K blk.37.ffn_down_shexp.weight q3_K +blk.38.attn_k_b.weight q4_0 blk.38.attn_output.weight q3_K blk.38.ffn_down_exps.weight q3_K blk.38.ffn_down_shexp.weight q3_K +blk.39.attn_k_b.weight q4_0 blk.39.attn_output.weight q3_K blk.39.ffn_down_exps.weight q3_K blk.39.ffn_down_shexp.weight q3_K +blk.40.attn_k_b.weight q4_0 blk.40.attn_output.weight q3_K blk.40.ffn_down_exps.weight q3_K blk.40.ffn_down_shexp.weight q3_K +blk.41.attn_k_b.weight q4_0 blk.41.attn_output.weight q3_K blk.41.ffn_down_exps.weight q3_K blk.41.ffn_down_shexp.weight q3_K +blk.42.attn_k_b.weight q4_0 blk.42.attn_output.weight q3_K blk.42.ffn_down_exps.weight q3_K blk.42.ffn_down_shexp.weight q3_K +blk.43.attn_k_b.weight q4_0 blk.43.attn_output.weight q3_K blk.43.ffn_down_exps.weight q3_K blk.43.ffn_down_shexp.weight q3_K +blk.44.attn_k_b.weight q4_0 blk.44.attn_output.weight q3_K blk.44.ffn_down_exps.weight q3_K blk.44.ffn_down_shexp.weight q3_K +blk.45.attn_k_b.weight q4_0 blk.45.attn_output.weight q3_K blk.45.ffn_down_exps.weight q3_K blk.45.ffn_down_shexp.weight q3_K +blk.46.attn_k_b.weight q4_0 blk.46.attn_output.weight q3_K blk.46.ffn_down_exps.weight q3_K blk.46.ffn_down_shexp.weight q3_K +blk.47.attn_k_b.weight q4_0 blk.47.attn_output.weight q3_K blk.47.ffn_down_exps.weight q3_K blk.47.ffn_down_shexp.weight q3_K +blk.48.attn_k_b.weight q4_0 blk.48.attn_output.weight q3_K blk.48.ffn_down_exps.weight q3_K blk.48.ffn_down_shexp.weight q3_K +blk.49.attn_k_b.weight q4_0 blk.49.attn_output.weight q3_K blk.49.ffn_down_exps.weight q3_K blk.49.ffn_down_shexp.weight q3_K +blk.50.attn_k_b.weight q4_0 blk.50.attn_output.weight q3_K blk.50.ffn_down_exps.weight q3_K blk.50.ffn_down_shexp.weight q3_K +blk.51.attn_k_b.weight q4_0 blk.51.attn_output.weight q3_K blk.51.ffn_down_exps.weight q3_K blk.51.ffn_down_shexp.weight q3_K +blk.52.attn_k_b.weight q4_0 blk.52.attn_output.weight q3_K blk.52.ffn_down_exps.weight q3_K blk.52.ffn_down_shexp.weight q3_K +blk.53.attn_k_b.weight q4_0 blk.53.attn_output.weight q3_K blk.53.ffn_down_exps.weight q3_K blk.53.ffn_down_shexp.weight q3_K +blk.54.attn_k_b.weight q4_0 blk.54.attn_output.weight q3_K blk.54.ffn_down_exps.weight q3_K blk.54.ffn_down_shexp.weight q3_K +blk.55.attn_k_b.weight q4_0 blk.55.attn_output.weight q3_K blk.55.ffn_down_exps.weight q3_K blk.55.ffn_down_shexp.weight q3_K +blk.56.attn_k_b.weight q4_0 blk.56.attn_output.weight q3_K blk.56.ffn_down_exps.weight q3_K blk.56.ffn_down_shexp.weight q3_K +blk.57.attn_k_b.weight q4_0 blk.57.attn_output.weight q3_K blk.57.ffn_down_exps.weight q3_K blk.57.ffn_down_shexp.weight q3_K +blk.58.attn_k_b.weight q4_0 blk.58.attn_output.weight q3_K blk.58.ffn_down_exps.weight q3_K blk.58.ffn_down_shexp.weight q3_K +blk.59.attn_k_b.weight q4_0 blk.59.attn_output.weight q3_K blk.59.ffn_down_exps.weight q3_K blk.59.ffn_down_shexp.weight q3_K +blk.60.attn_k_b.weight q4_0 blk.60.attn_output.weight q3_K blk.60.ffn_down_exps.weight q3_K blk.60.ffn_down_shexp.weight q3_K [Q3_K_S] q3_K output.weight q6_K +blk.0.attn_k_b.weight q4_0 +blk.1.attn_k_b.weight q4_0 +blk.2.attn_k_b.weight q4_0 +blk.3.attn_k_b.weight q4_0 +blk.4.attn_k_b.weight q4_0 +blk.5.attn_k_b.weight q4_0 +blk.6.attn_k_b.weight q4_0 +blk.7.attn_k_b.weight q4_0 +blk.8.attn_k_b.weight q4_0 +blk.9.attn_k_b.weight q4_0 +blk.10.attn_k_b.weight q4_0 +blk.11.attn_k_b.weight q4_0 +blk.12.attn_k_b.weight q4_0 +blk.13.attn_k_b.weight q4_0 +blk.14.attn_k_b.weight q4_0 +blk.15.attn_k_b.weight q4_0 +blk.16.attn_k_b.weight q4_0 +blk.17.attn_k_b.weight q4_0 +blk.18.attn_k_b.weight q4_0 +blk.19.attn_k_b.weight q4_0 +blk.20.attn_k_b.weight q4_0 +blk.21.attn_k_b.weight q4_0 +blk.22.attn_k_b.weight q4_0 +blk.23.attn_k_b.weight q4_0 +blk.24.attn_k_b.weight q4_0 +blk.25.attn_k_b.weight q4_0 +blk.26.attn_k_b.weight q4_0 +blk.27.attn_k_b.weight q4_0 +blk.28.attn_k_b.weight q4_0 +blk.29.attn_k_b.weight q4_0 +blk.30.attn_k_b.weight q4_0 +blk.31.attn_k_b.weight q4_0 +blk.32.attn_k_b.weight q4_0 +blk.33.attn_k_b.weight q4_0 +blk.34.attn_k_b.weight q4_0 +blk.35.attn_k_b.weight q4_0 +blk.36.attn_k_b.weight q4_0 +blk.37.attn_k_b.weight q4_0 +blk.38.attn_k_b.weight q4_0 +blk.39.attn_k_b.weight q4_0 +blk.40.attn_k_b.weight q4_0 +blk.41.attn_k_b.weight q4_0 +blk.42.attn_k_b.weight q4_0 +blk.43.attn_k_b.weight q4_0 +blk.44.attn_k_b.weight q4_0 +blk.45.attn_k_b.weight q4_0 +blk.46.attn_k_b.weight q4_0 +blk.47.attn_k_b.weight q4_0 +blk.48.attn_k_b.weight q4_0 +blk.49.attn_k_b.weight q4_0 +blk.50.attn_k_b.weight q4_0 +blk.51.attn_k_b.weight q4_0 +blk.52.attn_k_b.weight q4_0 +blk.53.attn_k_b.weight q4_0 +blk.54.attn_k_b.weight q4_0 +blk.55.attn_k_b.weight q4_0 +blk.56.attn_k_b.weight q4_0 +blk.57.attn_k_b.weight q4_0 +blk.58.attn_k_b.weight q4_0 +blk.59.attn_k_b.weight q4_0 +blk.60.attn_k_b.weight q4_0 [Q3_K_M] q3_K output.weight q6_K +blk.0.attn_k_b.weight q4_0 blk.0.attn_output.weight q4_K blk.0.ffn_down.weight q5_K +blk.1.attn_k_b.weight q4_0 blk.1.attn_output.weight q4_K blk.1.ffn_down.weight q5_K +blk.2.attn_k_b.weight q4_0 blk.2.attn_output.weight q4_K blk.2.ffn_down.weight q5_K +blk.3.attn_k_b.weight q4_0 blk.3.attn_output.weight q4_K blk.3.ffn_down_exps.weight q4_K blk.3.ffn_down_shexp.weight q4_K +blk.4.attn_k_b.weight q4_0 blk.4.attn_output.weight q4_K blk.4.ffn_down_exps.weight q4_K blk.4.ffn_down_shexp.weight q4_K +blk.5.attn_k_b.weight q4_0 blk.5.attn_output.weight q4_K blk.5.ffn_down_exps.weight q4_K blk.5.ffn_down_shexp.weight q4_K +blk.6.attn_k_b.weight q4_0 blk.6.attn_output.weight q4_K blk.6.ffn_down_exps.weight q4_K blk.6.ffn_down_shexp.weight q4_K +blk.7.attn_k_b.weight q4_0 blk.7.attn_output.weight q4_K blk.7.ffn_down_exps.weight q4_K blk.7.ffn_down_shexp.weight q4_K +blk.8.attn_k_b.weight q4_0 blk.8.attn_output.weight q4_K blk.8.ffn_down_exps.weight q4_K blk.8.ffn_down_shexp.weight q4_K +blk.9.attn_k_b.weight q4_0 blk.9.attn_output.weight q4_K blk.9.ffn_down_exps.weight q4_K blk.9.ffn_down_shexp.weight q4_K +blk.10.attn_k_b.weight q4_0 blk.10.attn_output.weight q4_K blk.10.ffn_down_exps.weight q4_K blk.10.ffn_down_shexp.weight q4_K +blk.11.attn_k_b.weight q4_0 blk.11.attn_output.weight q4_K blk.11.ffn_down_exps.weight q4_K blk.11.ffn_down_shexp.weight q4_K +blk.12.attn_k_b.weight q4_0 blk.12.attn_output.weight q4_K blk.12.ffn_down_exps.weight q4_K blk.12.ffn_down_shexp.weight q4_K +blk.13.attn_k_b.weight q4_0 blk.13.attn_output.weight q4_K blk.13.ffn_down_exps.weight q4_K blk.13.ffn_down_shexp.weight q4_K +blk.14.attn_k_b.weight q4_0 blk.14.attn_output.weight q4_K blk.14.ffn_down_exps.weight q4_K blk.14.ffn_down_shexp.weight q4_K +blk.15.attn_k_b.weight q4_0 blk.15.attn_output.weight q4_K blk.15.ffn_down_exps.weight q4_K blk.15.ffn_down_shexp.weight q4_K +blk.16.attn_k_b.weight q4_0 blk.16.attn_output.weight q4_K blk.16.ffn_down_exps.weight q4_K blk.16.ffn_down_shexp.weight q4_K +blk.17.attn_k_b.weight q4_0 blk.17.attn_output.weight q4_K blk.17.ffn_down_exps.weight q4_K blk.17.ffn_down_shexp.weight q4_K +blk.18.attn_k_b.weight q4_0 blk.18.attn_output.weight q4_K blk.18.ffn_down_exps.weight q4_K blk.18.ffn_down_shexp.weight q4_K +blk.19.attn_k_b.weight q4_0 blk.19.attn_output.weight q4_K blk.19.ffn_down_exps.weight q4_K blk.19.ffn_down_shexp.weight q4_K +blk.20.attn_k_b.weight q4_0 blk.20.attn_output.weight q4_K blk.20.ffn_down_exps.weight q4_K blk.20.ffn_down_shexp.weight q4_K +blk.21.attn_k_b.weight q4_0 blk.21.attn_output.weight q4_K blk.21.ffn_down_exps.weight q4_K blk.21.ffn_down_shexp.weight q4_K +blk.22.attn_k_b.weight q4_0 blk.22.attn_output.weight q4_K blk.22.ffn_down_exps.weight q4_K blk.22.ffn_down_shexp.weight q4_K +blk.23.attn_k_b.weight q4_0 blk.23.attn_output.weight q4_K blk.23.ffn_down_exps.weight q4_K blk.23.ffn_down_shexp.weight q4_K +blk.24.attn_k_b.weight q4_0 blk.24.attn_output.weight q4_K blk.24.ffn_down_exps.weight q4_K blk.24.ffn_down_shexp.weight q4_K +blk.25.attn_k_b.weight q4_0 blk.25.attn_output.weight q4_K blk.25.ffn_down_exps.weight q4_K blk.25.ffn_down_shexp.weight q4_K +blk.26.attn_k_b.weight q4_0 blk.26.attn_output.weight q4_K blk.26.ffn_down_exps.weight q4_K blk.26.ffn_down_shexp.weight q4_K +blk.27.attn_k_b.weight q4_0 blk.27.attn_output.weight q4_K blk.27.ffn_down_exps.weight q4_K blk.27.ffn_down_shexp.weight q4_K +blk.28.attn_k_b.weight q4_0 blk.28.attn_output.weight q4_K blk.28.ffn_down_exps.weight q4_K blk.28.ffn_down_shexp.weight q4_K +blk.29.attn_k_b.weight q4_0 blk.29.attn_output.weight q4_K blk.29.ffn_down_exps.weight q4_K blk.29.ffn_down_shexp.weight q4_K +blk.30.attn_k_b.weight q4_0 blk.30.attn_output.weight q4_K blk.30.ffn_down_exps.weight q4_K blk.30.ffn_down_shexp.weight q4_K +blk.31.attn_k_b.weight q4_0 blk.31.attn_output.weight q4_K blk.31.ffn_down_exps.weight q4_K blk.31.ffn_down_shexp.weight q4_K +blk.32.attn_k_b.weight q4_0 blk.32.attn_output.weight q4_K blk.32.ffn_down_exps.weight q4_K blk.32.ffn_down_shexp.weight q4_K +blk.33.attn_k_b.weight q4_0 blk.33.attn_output.weight q4_K blk.33.ffn_down_exps.weight q4_K blk.33.ffn_down_shexp.weight q4_K +blk.34.attn_k_b.weight q4_0 blk.34.attn_output.weight q4_K blk.34.ffn_down_exps.weight q4_K blk.34.ffn_down_shexp.weight q4_K +blk.35.attn_k_b.weight q4_0 blk.35.attn_output.weight q4_K blk.35.ffn_down_exps.weight q4_K blk.35.ffn_down_shexp.weight q4_K +blk.36.attn_k_b.weight q4_0 blk.36.attn_output.weight q4_K blk.36.ffn_down_exps.weight q4_K blk.36.ffn_down_shexp.weight q4_K +blk.37.attn_k_b.weight q4_0 blk.37.attn_output.weight q4_K blk.37.ffn_down_exps.weight q4_K blk.37.ffn_down_shexp.weight q4_K +blk.38.attn_k_b.weight q4_0 blk.38.attn_output.weight q4_K blk.38.ffn_down_exps.weight q4_K blk.38.ffn_down_shexp.weight q4_K +blk.39.attn_k_b.weight q4_0 blk.39.attn_output.weight q4_K blk.39.ffn_down_exps.weight q4_K blk.39.ffn_down_shexp.weight q4_K +blk.40.attn_k_b.weight q4_0 blk.40.attn_output.weight q4_K blk.40.ffn_down_exps.weight q4_K blk.40.ffn_down_shexp.weight q4_K +blk.41.attn_k_b.weight q4_0 blk.41.attn_output.weight q4_K blk.41.ffn_down_exps.weight q4_K blk.41.ffn_down_shexp.weight q4_K +blk.42.attn_k_b.weight q4_0 blk.42.attn_output.weight q4_K blk.42.ffn_down_exps.weight q4_K blk.42.ffn_down_shexp.weight q4_K +blk.43.attn_k_b.weight q4_0 blk.43.attn_output.weight q4_K blk.43.ffn_down_exps.weight q4_K blk.43.ffn_down_shexp.weight q4_K +blk.44.attn_k_b.weight q4_0 blk.44.attn_output.weight q4_K blk.44.ffn_down_exps.weight q4_K blk.44.ffn_down_shexp.weight q4_K +blk.45.attn_k_b.weight q4_0 blk.45.attn_output.weight q4_K blk.45.ffn_down_exps.weight q4_K blk.45.ffn_down_shexp.weight q4_K +blk.46.attn_k_b.weight q4_0 blk.46.attn_output.weight q4_K blk.46.ffn_down_exps.weight q4_K blk.46.ffn_down_shexp.weight q4_K +blk.47.attn_k_b.weight q4_0 blk.47.attn_output.weight q4_K blk.47.ffn_down_exps.weight q4_K blk.47.ffn_down_shexp.weight q4_K +blk.48.attn_k_b.weight q4_0 blk.48.attn_output.weight q4_K blk.48.ffn_down_exps.weight q4_K blk.48.ffn_down_shexp.weight q4_K +blk.49.attn_k_b.weight q4_0 blk.49.attn_output.weight q4_K blk.49.ffn_down_exps.weight q4_K blk.49.ffn_down_shexp.weight q4_K +blk.50.attn_k_b.weight q4_0 blk.50.attn_output.weight q4_K blk.50.ffn_down_exps.weight q4_K blk.50.ffn_down_shexp.weight q4_K +blk.51.attn_k_b.weight q4_0 blk.51.attn_output.weight q4_K blk.51.ffn_down_exps.weight q4_K blk.51.ffn_down_shexp.weight q4_K +blk.52.attn_k_b.weight q4_0 blk.52.attn_output.weight q4_K blk.52.ffn_down_exps.weight q4_K blk.52.ffn_down_shexp.weight q4_K +blk.53.attn_k_b.weight q4_0 blk.53.attn_output.weight q4_K blk.53.ffn_down_exps.weight q4_K blk.53.ffn_down_shexp.weight q4_K +blk.54.attn_k_b.weight q4_0 blk.54.attn_output.weight q4_K blk.54.ffn_down_exps.weight q4_K blk.54.ffn_down_shexp.weight q4_K +blk.55.attn_k_b.weight q4_0 blk.55.attn_output.weight q4_K blk.55.ffn_down_exps.weight q4_K blk.55.ffn_down_shexp.weight q4_K +blk.56.attn_k_b.weight q4_0 blk.56.attn_output.weight q4_K blk.56.ffn_down_exps.weight q4_K blk.56.ffn_down_shexp.weight q4_K +blk.57.attn_k_b.weight q4_0 blk.57.attn_output.weight q4_K blk.57.ffn_down_exps.weight q4_K blk.57.ffn_down_shexp.weight q4_K +blk.58.attn_k_b.weight q4_0 blk.58.attn_output.weight q4_K blk.58.ffn_down_exps.weight q4_K blk.58.ffn_down_shexp.weight q4_K +blk.59.attn_k_b.weight q4_0 blk.59.attn_output.weight q4_K blk.59.ffn_down_exps.weight q4_K blk.59.ffn_down_shexp.weight q4_K +blk.60.attn_k_b.weight q4_0 blk.60.attn_output.weight q4_K blk.60.ffn_down_exps.weight q4_K blk.60.ffn_down_shexp.weight q4_K [Q3_K_L] q3_K output.weight q6_K +blk.0.attn_k_b.weight q4_0 blk.0.attn_output.weight q5_K blk.0.ffn_down.weight q5_K +blk.1.attn_k_b.weight q4_0 blk.1.attn_output.weight q5_K blk.1.ffn_down.weight q5_K +blk.2.attn_k_b.weight q4_0 blk.2.attn_output.weight q5_K blk.2.ffn_down.weight q5_K +blk.3.attn_k_b.weight q4_0 blk.3.attn_output.weight q5_K blk.3.ffn_down_exps.weight q5_K blk.3.ffn_down_shexp.weight q5_K +blk.4.attn_k_b.weight q4_0 blk.4.attn_output.weight q5_K blk.4.ffn_down_exps.weight q5_K blk.4.ffn_down_shexp.weight q5_K +blk.5.attn_k_b.weight q4_0 blk.5.attn_output.weight q5_K blk.5.ffn_down_exps.weight q5_K blk.5.ffn_down_shexp.weight q5_K +blk.6.attn_k_b.weight q4_0 blk.6.attn_output.weight q5_K blk.6.ffn_down_exps.weight q5_K blk.6.ffn_down_shexp.weight q5_K +blk.7.attn_k_b.weight q4_0 blk.7.attn_output.weight q5_K blk.7.ffn_down_exps.weight q5_K blk.7.ffn_down_shexp.weight q5_K +blk.8.attn_k_b.weight q4_0 blk.8.attn_output.weight q5_K blk.8.ffn_down_exps.weight q5_K blk.8.ffn_down_shexp.weight q5_K +blk.9.attn_k_b.weight q4_0 blk.9.attn_output.weight q5_K blk.9.ffn_down_exps.weight q5_K blk.9.ffn_down_shexp.weight q5_K +blk.10.attn_k_b.weight q4_0 blk.10.attn_output.weight q5_K blk.10.ffn_down_exps.weight q5_K blk.10.ffn_down_shexp.weight q5_K +blk.11.attn_k_b.weight q4_0 blk.11.attn_output.weight q5_K blk.11.ffn_down_exps.weight q5_K blk.11.ffn_down_shexp.weight q5_K +blk.12.attn_k_b.weight q4_0 blk.12.attn_output.weight q5_K blk.12.ffn_down_exps.weight q5_K blk.12.ffn_down_shexp.weight q5_K +blk.13.attn_k_b.weight q4_0 blk.13.attn_output.weight q5_K blk.13.ffn_down_exps.weight q5_K blk.13.ffn_down_shexp.weight q5_K +blk.14.attn_k_b.weight q4_0 blk.14.attn_output.weight q5_K blk.14.ffn_down_exps.weight q5_K blk.14.ffn_down_shexp.weight q5_K +blk.15.attn_k_b.weight q4_0 blk.15.attn_output.weight q5_K blk.15.ffn_down_exps.weight q5_K blk.15.ffn_down_shexp.weight q5_K +blk.16.attn_k_b.weight q4_0 blk.16.attn_output.weight q5_K blk.16.ffn_down_exps.weight q5_K blk.16.ffn_down_shexp.weight q5_K +blk.17.attn_k_b.weight q4_0 blk.17.attn_output.weight q5_K blk.17.ffn_down_exps.weight q5_K blk.17.ffn_down_shexp.weight q5_K +blk.18.attn_k_b.weight q4_0 blk.18.attn_output.weight q5_K blk.18.ffn_down_exps.weight q5_K blk.18.ffn_down_shexp.weight q5_K +blk.19.attn_k_b.weight q4_0 blk.19.attn_output.weight q5_K blk.19.ffn_down_exps.weight q5_K blk.19.ffn_down_shexp.weight q5_K +blk.20.attn_k_b.weight q4_0 blk.20.attn_output.weight q5_K blk.20.ffn_down_exps.weight q5_K blk.20.ffn_down_shexp.weight q5_K +blk.21.attn_k_b.weight q4_0 blk.21.attn_output.weight q5_K blk.21.ffn_down_exps.weight q5_K blk.21.ffn_down_shexp.weight q5_K +blk.22.attn_k_b.weight q4_0 blk.22.attn_output.weight q5_K blk.22.ffn_down_exps.weight q5_K blk.22.ffn_down_shexp.weight q5_K +blk.23.attn_k_b.weight q4_0 blk.23.attn_output.weight q5_K blk.23.ffn_down_exps.weight q5_K blk.23.ffn_down_shexp.weight q5_K +blk.24.attn_k_b.weight q4_0 blk.24.attn_output.weight q5_K blk.24.ffn_down_exps.weight q5_K blk.24.ffn_down_shexp.weight q5_K +blk.25.attn_k_b.weight q4_0 blk.25.attn_output.weight q5_K blk.25.ffn_down_exps.weight q5_K blk.25.ffn_down_shexp.weight q5_K +blk.26.attn_k_b.weight q4_0 blk.26.attn_output.weight q5_K blk.26.ffn_down_exps.weight q5_K blk.26.ffn_down_shexp.weight q5_K +blk.27.attn_k_b.weight q4_0 blk.27.attn_output.weight q5_K blk.27.ffn_down_exps.weight q5_K blk.27.ffn_down_shexp.weight q5_K +blk.28.attn_k_b.weight q4_0 blk.28.attn_output.weight q5_K blk.28.ffn_down_exps.weight q5_K blk.28.ffn_down_shexp.weight q5_K +blk.29.attn_k_b.weight q4_0 blk.29.attn_output.weight q5_K blk.29.ffn_down_exps.weight q5_K blk.29.ffn_down_shexp.weight q5_K +blk.30.attn_k_b.weight q4_0 blk.30.attn_output.weight q5_K blk.30.ffn_down_exps.weight q5_K blk.30.ffn_down_shexp.weight q5_K +blk.31.attn_k_b.weight q4_0 blk.31.attn_output.weight q5_K blk.31.ffn_down_exps.weight q5_K blk.31.ffn_down_shexp.weight q5_K +blk.32.attn_k_b.weight q4_0 blk.32.attn_output.weight q5_K blk.32.ffn_down_exps.weight q5_K blk.32.ffn_down_shexp.weight q5_K +blk.33.attn_k_b.weight q4_0 blk.33.attn_output.weight q5_K blk.33.ffn_down_exps.weight q5_K blk.33.ffn_down_shexp.weight q5_K +blk.34.attn_k_b.weight q4_0 blk.34.attn_output.weight q5_K blk.34.ffn_down_exps.weight q5_K blk.34.ffn_down_shexp.weight q5_K +blk.35.attn_k_b.weight q4_0 blk.35.attn_output.weight q5_K blk.35.ffn_down_exps.weight q5_K blk.35.ffn_down_shexp.weight q5_K +blk.36.attn_k_b.weight q4_0 blk.36.attn_output.weight q5_K blk.36.ffn_down_exps.weight q5_K blk.36.ffn_down_shexp.weight q5_K +blk.37.attn_k_b.weight q4_0 blk.37.attn_output.weight q5_K blk.37.ffn_down_exps.weight q5_K blk.37.ffn_down_shexp.weight q5_K +blk.38.attn_k_b.weight q4_0 blk.38.attn_output.weight q5_K blk.38.ffn_down_exps.weight q5_K blk.38.ffn_down_shexp.weight q5_K +blk.39.attn_k_b.weight q4_0 blk.39.attn_output.weight q5_K blk.39.ffn_down_exps.weight q5_K blk.39.ffn_down_shexp.weight q5_K +blk.40.attn_k_b.weight q4_0 blk.40.attn_output.weight q5_K blk.40.ffn_down_exps.weight q5_K blk.40.ffn_down_shexp.weight q5_K +blk.41.attn_k_b.weight q4_0 blk.41.attn_output.weight q5_K blk.41.ffn_down_exps.weight q5_K blk.41.ffn_down_shexp.weight q5_K +blk.42.attn_k_b.weight q4_0 blk.42.attn_output.weight q5_K blk.42.ffn_down_exps.weight q5_K blk.42.ffn_down_shexp.weight q5_K +blk.43.attn_k_b.weight q4_0 blk.43.attn_output.weight q5_K blk.43.ffn_down_exps.weight q5_K blk.43.ffn_down_shexp.weight q5_K +blk.44.attn_k_b.weight q4_0 blk.44.attn_output.weight q5_K blk.44.ffn_down_exps.weight q5_K blk.44.ffn_down_shexp.weight q5_K +blk.45.attn_k_b.weight q4_0 blk.45.attn_output.weight q5_K blk.45.ffn_down_exps.weight q5_K blk.45.ffn_down_shexp.weight q5_K +blk.46.attn_k_b.weight q4_0 blk.46.attn_output.weight q5_K blk.46.ffn_down_exps.weight q5_K blk.46.ffn_down_shexp.weight q5_K +blk.47.attn_k_b.weight q4_0 blk.47.attn_output.weight q5_K blk.47.ffn_down_exps.weight q5_K blk.47.ffn_down_shexp.weight q5_K +blk.48.attn_k_b.weight q4_0 blk.48.attn_output.weight q5_K blk.48.ffn_down_exps.weight q5_K blk.48.ffn_down_shexp.weight q5_K +blk.49.attn_k_b.weight q4_0 blk.49.attn_output.weight q5_K blk.49.ffn_down_exps.weight q5_K blk.49.ffn_down_shexp.weight q5_K +blk.50.attn_k_b.weight q4_0 blk.50.attn_output.weight q5_K blk.50.ffn_down_exps.weight q5_K blk.50.ffn_down_shexp.weight q5_K +blk.51.attn_k_b.weight q4_0 blk.51.attn_output.weight q5_K blk.51.ffn_down_exps.weight q5_K blk.51.ffn_down_shexp.weight q5_K +blk.52.attn_k_b.weight q4_0 blk.52.attn_output.weight q5_K blk.52.ffn_down_exps.weight q5_K blk.52.ffn_down_shexp.weight q5_K +blk.53.attn_k_b.weight q4_0 blk.53.attn_output.weight q5_K blk.53.ffn_down_exps.weight q5_K blk.53.ffn_down_shexp.weight q5_K +blk.54.attn_k_b.weight q4_0 blk.54.attn_output.weight q5_K blk.54.ffn_down_exps.weight q5_K blk.54.ffn_down_shexp.weight q5_K +blk.55.attn_k_b.weight q4_0 blk.55.attn_output.weight q5_K blk.55.ffn_down_exps.weight q5_K blk.55.ffn_down_shexp.weight q5_K +blk.56.attn_k_b.weight q4_0 blk.56.attn_output.weight q5_K blk.56.ffn_down_exps.weight q5_K blk.56.ffn_down_shexp.weight q5_K +blk.57.attn_k_b.weight q4_0 blk.57.attn_output.weight q5_K blk.57.ffn_down_exps.weight q5_K blk.57.ffn_down_shexp.weight q5_K +blk.58.attn_k_b.weight q4_0 blk.58.attn_output.weight q5_K blk.58.ffn_down_exps.weight q5_K blk.58.ffn_down_shexp.weight q5_K +blk.59.attn_k_b.weight q4_0 blk.59.attn_output.weight q5_K blk.59.ffn_down_exps.weight q5_K blk.59.ffn_down_shexp.weight q5_K +blk.60.attn_k_b.weight q4_0 blk.60.attn_output.weight q5_K blk.60.ffn_down_exps.weight q5_K blk.60.ffn_down_shexp.weight q5_K [Q4_K_S] q4_K output.weight q6_K +blk.0.attn_k_b.weight q5_0 blk.0.ffn_down.weight q5_K +blk.1.attn_k_b.weight q5_0 blk.1.ffn_down.weight q5_K +blk.2.attn_k_b.weight q5_0 blk.2.ffn_down.weight q5_K +blk.3.attn_k_b.weight q5_0 blk.3.ffn_down_exps.weight q5_K blk.3.ffn_down_shexp.weight q5_K +blk.4.attn_k_b.weight q5_0 blk.4.ffn_down_exps.weight q5_K blk.4.ffn_down_shexp.weight q5_K +blk.5.attn_k_b.weight q5_0 blk.5.ffn_down_exps.weight q5_K blk.5.ffn_down_shexp.weight q5_K +blk.6.attn_k_b.weight q5_0 blk.6.ffn_down_exps.weight q5_K blk.6.ffn_down_shexp.weight q5_K +blk.7.attn_k_b.weight q5_0 +blk.8.attn_k_b.weight q5_0 +blk.9.attn_k_b.weight q5_0 +blk.10.attn_k_b.weight q5_0 +blk.11.attn_k_b.weight q5_0 +blk.12.attn_k_b.weight q5_0 +blk.13.attn_k_b.weight q5_0 +blk.14.attn_k_b.weight q5_0 +blk.15.attn_k_b.weight q5_0 +blk.16.attn_k_b.weight q5_0 +blk.17.attn_k_b.weight q5_0 +blk.18.attn_k_b.weight q5_0 +blk.19.attn_k_b.weight q5_0 +blk.20.attn_k_b.weight q5_0 +blk.21.attn_k_b.weight q5_0 +blk.22.attn_k_b.weight q5_0 +blk.23.attn_k_b.weight q5_0 +blk.24.attn_k_b.weight q5_0 +blk.25.attn_k_b.weight q5_0 +blk.26.attn_k_b.weight q5_0 +blk.27.attn_k_b.weight q5_0 +blk.28.attn_k_b.weight q5_0 +blk.29.attn_k_b.weight q5_0 +blk.30.attn_k_b.weight q5_0 +blk.31.attn_k_b.weight q5_0 +blk.32.attn_k_b.weight q5_0 +blk.33.attn_k_b.weight q5_0 +blk.34.attn_k_b.weight q5_0 +blk.35.attn_k_b.weight q5_0 +blk.36.attn_k_b.weight q5_0 +blk.37.attn_k_b.weight q5_0 +blk.38.attn_k_b.weight q5_0 +blk.39.attn_k_b.weight q5_0 +blk.40.attn_k_b.weight q5_0 +blk.41.attn_k_b.weight q5_0 +blk.42.attn_k_b.weight q5_0 +blk.43.attn_k_b.weight q5_0 +blk.44.attn_k_b.weight q5_0 +blk.45.attn_k_b.weight q5_0 +blk.46.attn_k_b.weight q5_0 +blk.47.attn_k_b.weight q5_0 +blk.48.attn_k_b.weight q5_0 +blk.49.attn_k_b.weight q5_0 +blk.50.attn_k_b.weight q5_0 +blk.51.attn_k_b.weight q5_0 +blk.52.attn_k_b.weight q5_0 +blk.53.attn_k_b.weight q5_0 +blk.54.attn_k_b.weight q5_0 +blk.55.attn_k_b.weight q5_0 +blk.56.attn_k_b.weight q5_0 +blk.57.attn_k_b.weight q5_0 +blk.58.attn_k_b.weight q5_0 +blk.59.attn_k_b.weight q5_0 +blk.60.attn_k_b.weight q5_0 [Q4_K_M] q4_K output.weight q6_K +blk.0.attn_k_b.weight q5_0 blk.0.ffn_down.weight q6_K +blk.1.attn_k_b.weight q5_0 blk.1.ffn_down.weight q6_K +blk.2.attn_k_b.weight q5_0 blk.2.ffn_down.weight q6_K +blk.3.attn_k_b.weight q5_0 blk.3.ffn_down_exps.weight q6_K blk.3.ffn_down_shexp.weight q6_K +blk.4.attn_k_b.weight q5_0 blk.4.ffn_down_exps.weight q6_K blk.4.ffn_down_shexp.weight q6_K +blk.5.attn_k_b.weight q5_0 blk.5.ffn_down_exps.weight q6_K blk.5.ffn_down_shexp.weight q6_K +blk.6.attn_k_b.weight q5_0 blk.6.ffn_down_exps.weight q6_K blk.6.ffn_down_shexp.weight q6_K +blk.7.attn_k_b.weight q5_0 +blk.8.attn_k_b.weight q5_0 +blk.9.attn_k_b.weight q5_0 blk.9.ffn_down_exps.weight q6_K blk.9.ffn_down_shexp.weight q6_K +blk.10.attn_k_b.weight q5_0 +blk.11.attn_k_b.weight q5_0 +blk.12.attn_k_b.weight q5_0 blk.12.ffn_down_exps.weight q6_K blk.12.ffn_down_shexp.weight q6_K +blk.13.attn_k_b.weight q5_0 +blk.14.attn_k_b.weight q5_0 +blk.15.attn_k_b.weight q5_0 blk.15.ffn_down_exps.weight q6_K blk.15.ffn_down_shexp.weight q6_K +blk.16.attn_k_b.weight q5_0 +blk.17.attn_k_b.weight q5_0 +blk.18.attn_k_b.weight q5_0 blk.18.ffn_down_exps.weight q6_K blk.18.ffn_down_shexp.weight q6_K +blk.19.attn_k_b.weight q5_0 +blk.20.attn_k_b.weight q5_0 +blk.21.attn_k_b.weight q5_0 blk.21.ffn_down_exps.weight q6_K blk.21.ffn_down_shexp.weight q6_K +blk.22.attn_k_b.weight q5_0 +blk.23.attn_k_b.weight q5_0 +blk.24.attn_k_b.weight q5_0 blk.24.ffn_down_exps.weight q6_K blk.24.ffn_down_shexp.weight q6_K +blk.25.attn_k_b.weight q5_0 +blk.26.attn_k_b.weight q5_0 +blk.27.attn_k_b.weight q5_0 blk.27.ffn_down_exps.weight q6_K blk.27.ffn_down_shexp.weight q6_K +blk.28.attn_k_b.weight q5_0 +blk.29.attn_k_b.weight q5_0 +blk.30.attn_k_b.weight q5_0 blk.30.ffn_down_exps.weight q6_K blk.30.ffn_down_shexp.weight q6_K +blk.31.attn_k_b.weight q5_0 +blk.32.attn_k_b.weight q5_0 +blk.33.attn_k_b.weight q5_0 blk.33.ffn_down_exps.weight q6_K blk.33.ffn_down_shexp.weight q6_K +blk.34.attn_k_b.weight q5_0 +blk.35.attn_k_b.weight q5_0 +blk.36.attn_k_b.weight q5_0 blk.36.ffn_down_exps.weight q6_K blk.36.ffn_down_shexp.weight q6_K +blk.37.attn_k_b.weight q5_0 +blk.38.attn_k_b.weight q5_0 +blk.39.attn_k_b.weight q5_0 blk.39.ffn_down_exps.weight q6_K blk.39.ffn_down_shexp.weight q6_K +blk.40.attn_k_b.weight q5_0 +blk.41.attn_k_b.weight q5_0 +blk.42.attn_k_b.weight q5_0 blk.42.ffn_down_exps.weight q6_K blk.42.ffn_down_shexp.weight q6_K +blk.43.attn_k_b.weight q5_0 +blk.44.attn_k_b.weight q5_0 +blk.45.attn_k_b.weight q5_0 blk.45.ffn_down_exps.weight q6_K blk.45.ffn_down_shexp.weight q6_K +blk.46.attn_k_b.weight q5_0 +blk.47.attn_k_b.weight q5_0 +blk.48.attn_k_b.weight q5_0 blk.48.ffn_down_exps.weight q6_K blk.48.ffn_down_shexp.weight q6_K +blk.49.attn_k_b.weight q5_0 +blk.50.attn_k_b.weight q5_0 +blk.51.attn_k_b.weight q5_0 blk.51.ffn_down_exps.weight q6_K blk.51.ffn_down_shexp.weight q6_K +blk.52.attn_k_b.weight q5_0 +blk.53.attn_k_b.weight q5_0 blk.53.ffn_down_exps.weight q6_K blk.53.ffn_down_shexp.weight q6_K +blk.54.attn_k_b.weight q5_0 blk.54.ffn_down_exps.weight q6_K blk.54.ffn_down_shexp.weight q6_K +blk.55.attn_k_b.weight q5_0 blk.55.ffn_down_exps.weight q6_K blk.55.ffn_down_shexp.weight q6_K +blk.56.attn_k_b.weight q5_0 blk.56.ffn_down_exps.weight q6_K blk.56.ffn_down_shexp.weight q6_K +blk.57.attn_k_b.weight q5_0 blk.57.ffn_down_exps.weight q6_K blk.57.ffn_down_shexp.weight q6_K +blk.58.attn_k_b.weight q5_0 blk.58.ffn_down_exps.weight q6_K blk.58.ffn_down_shexp.weight q6_K +blk.59.attn_k_b.weight q5_0 blk.59.ffn_down_exps.weight q6_K blk.59.ffn_down_shexp.weight q6_K +blk.60.attn_k_b.weight q5_0 blk.60.ffn_down_exps.weight q6_K blk.60.ffn_down_shexp.weight q6_K [Q5_K_S] q5_K output.weight q6_K +blk.0.attn_k_b.weight q5_1 +blk.1.attn_k_b.weight q5_1 +blk.2.attn_k_b.weight q5_1 +blk.3.attn_k_b.weight q5_1 +blk.4.attn_k_b.weight q5_1 +blk.5.attn_k_b.weight q5_1 +blk.6.attn_k_b.weight q5_1 +blk.7.attn_k_b.weight q5_1 +blk.8.attn_k_b.weight q5_1 +blk.9.attn_k_b.weight q5_1 +blk.10.attn_k_b.weight q5_1 +blk.11.attn_k_b.weight q5_1 +blk.12.attn_k_b.weight q5_1 +blk.13.attn_k_b.weight q5_1 +blk.14.attn_k_b.weight q5_1 +blk.15.attn_k_b.weight q5_1 +blk.16.attn_k_b.weight q5_1 +blk.17.attn_k_b.weight q5_1 +blk.18.attn_k_b.weight q5_1 +blk.19.attn_k_b.weight q5_1 +blk.20.attn_k_b.weight q5_1 +blk.21.attn_k_b.weight q5_1 +blk.22.attn_k_b.weight q5_1 +blk.23.attn_k_b.weight q5_1 +blk.24.attn_k_b.weight q5_1 +blk.25.attn_k_b.weight q5_1 +blk.26.attn_k_b.weight q5_1 +blk.27.attn_k_b.weight q5_1 +blk.28.attn_k_b.weight q5_1 +blk.29.attn_k_b.weight q5_1 +blk.30.attn_k_b.weight q5_1 +blk.31.attn_k_b.weight q5_1 +blk.32.attn_k_b.weight q5_1 +blk.33.attn_k_b.weight q5_1 +blk.34.attn_k_b.weight q5_1 +blk.35.attn_k_b.weight q5_1 +blk.36.attn_k_b.weight q5_1 +blk.37.attn_k_b.weight q5_1 +blk.38.attn_k_b.weight q5_1 +blk.39.attn_k_b.weight q5_1 +blk.40.attn_k_b.weight q5_1 +blk.41.attn_k_b.weight q5_1 +blk.42.attn_k_b.weight q5_1 +blk.43.attn_k_b.weight q5_1 +blk.44.attn_k_b.weight q5_1 +blk.45.attn_k_b.weight q5_1 +blk.46.attn_k_b.weight q5_1 +blk.47.attn_k_b.weight q5_1 +blk.48.attn_k_b.weight q5_1 +blk.49.attn_k_b.weight q5_1 +blk.50.attn_k_b.weight q5_1 +blk.51.attn_k_b.weight q5_1 +blk.52.attn_k_b.weight q5_1 +blk.53.attn_k_b.weight q5_1 +blk.54.attn_k_b.weight q5_1 +blk.55.attn_k_b.weight q5_1 +blk.56.attn_k_b.weight q5_1 +blk.57.attn_k_b.weight q5_1 +blk.58.attn_k_b.weight q5_1 +blk.59.attn_k_b.weight q5_1 +blk.60.attn_k_b.weight q5_1 [Q5_K_M] q5_K output.weight q6_K +blk.0.attn_k_b.weight q5_1 blk.0.ffn_down.weight q6_K +blk.1.attn_k_b.weight q5_1 blk.1.ffn_down.weight q6_K +blk.2.attn_k_b.weight q5_1 blk.2.ffn_down.weight q6_K +blk.3.attn_k_b.weight q5_1 blk.3.ffn_down_exps.weight q6_K blk.3.ffn_down_shexp.weight q6_K +blk.4.attn_k_b.weight q5_1 blk.4.ffn_down_exps.weight q6_K blk.4.ffn_down_shexp.weight q6_K +blk.5.attn_k_b.weight q5_1 blk.5.ffn_down_exps.weight q6_K blk.5.ffn_down_shexp.weight q6_K +blk.6.attn_k_b.weight q5_1 blk.6.ffn_down_exps.weight q6_K blk.6.ffn_down_shexp.weight q6_K +blk.7.attn_k_b.weight q5_1 +blk.8.attn_k_b.weight q5_1 +blk.9.attn_k_b.weight q5_1 blk.9.ffn_down_exps.weight q6_K blk.9.ffn_down_shexp.weight q6_K +blk.10.attn_k_b.weight q5_1 +blk.11.attn_k_b.weight q5_1 +blk.12.attn_k_b.weight q5_1 blk.12.ffn_down_exps.weight q6_K blk.12.ffn_down_shexp.weight q6_K +blk.13.attn_k_b.weight q5_1 +blk.14.attn_k_b.weight q5_1 +blk.15.attn_k_b.weight q5_1 blk.15.ffn_down_exps.weight q6_K blk.15.ffn_down_shexp.weight q6_K +blk.16.attn_k_b.weight q5_1 +blk.17.attn_k_b.weight q5_1 +blk.18.attn_k_b.weight q5_1 blk.18.ffn_down_exps.weight q6_K blk.18.ffn_down_shexp.weight q6_K +blk.19.attn_k_b.weight q5_1 +blk.20.attn_k_b.weight q5_1 +blk.21.attn_k_b.weight q5_1 blk.21.ffn_down_exps.weight q6_K blk.21.ffn_down_shexp.weight q6_K +blk.22.attn_k_b.weight q5_1 +blk.23.attn_k_b.weight q5_1 +blk.24.attn_k_b.weight q5_1 blk.24.ffn_down_exps.weight q6_K blk.24.ffn_down_shexp.weight q6_K +blk.25.attn_k_b.weight q5_1 +blk.26.attn_k_b.weight q5_1 +blk.27.attn_k_b.weight q5_1 blk.27.ffn_down_exps.weight q6_K blk.27.ffn_down_shexp.weight q6_K +blk.28.attn_k_b.weight q5_1 +blk.29.attn_k_b.weight q5_1 +blk.30.attn_k_b.weight q5_1 blk.30.ffn_down_exps.weight q6_K blk.30.ffn_down_shexp.weight q6_K +blk.31.attn_k_b.weight q5_1 +blk.32.attn_k_b.weight q5_1 +blk.33.attn_k_b.weight q5_1 blk.33.ffn_down_exps.weight q6_K blk.33.ffn_down_shexp.weight q6_K +blk.34.attn_k_b.weight q5_1 +blk.35.attn_k_b.weight q5_1 +blk.36.attn_k_b.weight q5_1 blk.36.ffn_down_exps.weight q6_K blk.36.ffn_down_shexp.weight q6_K +blk.37.attn_k_b.weight q5_1 +blk.38.attn_k_b.weight q5_1 +blk.39.attn_k_b.weight q5_1 blk.39.ffn_down_exps.weight q6_K blk.39.ffn_down_shexp.weight q6_K +blk.40.attn_k_b.weight q5_1 +blk.41.attn_k_b.weight q5_1 +blk.42.attn_k_b.weight q5_1 blk.42.ffn_down_exps.weight q6_K blk.42.ffn_down_shexp.weight q6_K +blk.43.attn_k_b.weight q5_1 +blk.44.attn_k_b.weight q5_1 +blk.45.attn_k_b.weight q5_1 blk.45.ffn_down_exps.weight q6_K blk.45.ffn_down_shexp.weight q6_K +blk.46.attn_k_b.weight q5_1 +blk.47.attn_k_b.weight q5_1 +blk.48.attn_k_b.weight q5_1 blk.48.ffn_down_exps.weight q6_K blk.48.ffn_down_shexp.weight q6_K +blk.49.attn_k_b.weight q5_1 +blk.50.attn_k_b.weight q5_1 +blk.51.attn_k_b.weight q5_1 blk.51.ffn_down_exps.weight q6_K blk.51.ffn_down_shexp.weight q6_K +blk.52.attn_k_b.weight q5_1 +blk.53.attn_k_b.weight q5_1 blk.53.ffn_down_exps.weight q6_K blk.53.ffn_down_shexp.weight q6_K +blk.54.attn_k_b.weight q5_1 blk.54.ffn_down_exps.weight q6_K blk.54.ffn_down_shexp.weight q6_K +blk.55.attn_k_b.weight q5_1 blk.55.ffn_down_exps.weight q6_K blk.55.ffn_down_shexp.weight q6_K +blk.56.attn_k_b.weight q5_1 blk.56.ffn_down_exps.weight q6_K blk.56.ffn_down_shexp.weight q6_K +blk.57.attn_k_b.weight q5_1 blk.57.ffn_down_exps.weight q6_K blk.57.ffn_down_shexp.weight q6_K +blk.58.attn_k_b.weight q5_1 blk.58.ffn_down_exps.weight q6_K blk.58.ffn_down_shexp.weight q6_K +blk.59.attn_k_b.weight q5_1 blk.59.ffn_down_exps.weight q6_K blk.59.ffn_down_shexp.weight q6_K +blk.60.attn_k_b.weight q5_1 blk.60.ffn_down_exps.weight q6_K blk.60.ffn_down_shexp.weight q6_K [Q6_K] q6_K +blk.0.attn_k_b.weight q8_0 +blk.1.attn_k_b.weight q8_0 +blk.2.attn_k_b.weight q8_0 +blk.3.attn_k_b.weight q8_0 +blk.4.attn_k_b.weight q8_0 +blk.5.attn_k_b.weight q8_0 +blk.6.attn_k_b.weight q8_0 +blk.7.attn_k_b.weight q8_0 +blk.8.attn_k_b.weight q8_0 +blk.9.attn_k_b.weight q8_0 +blk.10.attn_k_b.weight q8_0 +blk.11.attn_k_b.weight q8_0 +blk.12.attn_k_b.weight q8_0 +blk.13.attn_k_b.weight q8_0 +blk.14.attn_k_b.weight q8_0 +blk.15.attn_k_b.weight q8_0 +blk.16.attn_k_b.weight q8_0 +blk.17.attn_k_b.weight q8_0 +blk.18.attn_k_b.weight q8_0 +blk.19.attn_k_b.weight q8_0 +blk.20.attn_k_b.weight q8_0 +blk.21.attn_k_b.weight q8_0 +blk.22.attn_k_b.weight q8_0 +blk.23.attn_k_b.weight q8_0 +blk.24.attn_k_b.weight q8_0 +blk.25.attn_k_b.weight q8_0 +blk.26.attn_k_b.weight q8_0 +blk.27.attn_k_b.weight q8_0 +blk.28.attn_k_b.weight q8_0 +blk.29.attn_k_b.weight q8_0 +blk.30.attn_k_b.weight q8_0 +blk.31.attn_k_b.weight q8_0 +blk.32.attn_k_b.weight q8_0 +blk.33.attn_k_b.weight q8_0 +blk.34.attn_k_b.weight q8_0 +blk.35.attn_k_b.weight q8_0 +blk.36.attn_k_b.weight q8_0 +blk.37.attn_k_b.weight q8_0 +blk.38.attn_k_b.weight q8_0 +blk.39.attn_k_b.weight q8_0 +blk.40.attn_k_b.weight q8_0 +blk.41.attn_k_b.weight q8_0 +blk.42.attn_k_b.weight q8_0 +blk.43.attn_k_b.weight q8_0 +blk.44.attn_k_b.weight q8_0 +blk.45.attn_k_b.weight q8_0 +blk.46.attn_k_b.weight q8_0 +blk.47.attn_k_b.weight q8_0 +blk.48.attn_k_b.weight q8_0 +blk.49.attn_k_b.weight q8_0 +blk.50.attn_k_b.weight q8_0 +blk.51.attn_k_b.weight q8_0 +blk.52.attn_k_b.weight q8_0 +blk.53.attn_k_b.weight q8_0 +blk.54.attn_k_b.weight q8_0 +blk.55.attn_k_b.weight q8_0 +blk.56.attn_k_b.weight q8_0 +blk.57.attn_k_b.weight q8_0 +blk.58.attn_k_b.weight q8_0 +blk.59.attn_k_b.weight q8_0 +blk.60.attn_k_b.weight q8_0 [IQ2_XXS] iq2_xxs output.weight q5_K token_embd.weight q2_K +blk.0.attn_k_b.weight iq4_nl blk.0.ffn_down.weight q2_K +blk.1.attn_k_b.weight iq4_nl blk.1.ffn_down.weight q2_K +blk.2.attn_k_b.weight iq4_nl blk.2.ffn_down.weight q2_K +blk.3.attn_k_b.weight iq4_nl blk.3.ffn_down_exps.weight q2_K blk.3.ffn_down_shexp.weight q2_K +blk.4.attn_k_b.weight iq4_nl blk.4.ffn_down_exps.weight q2_K blk.4.ffn_down_shexp.weight q2_K +blk.5.attn_k_b.weight iq4_nl +blk.6.attn_k_b.weight iq4_nl +blk.7.attn_k_b.weight iq4_nl +blk.8.attn_k_b.weight iq4_nl +blk.9.attn_k_b.weight iq4_nl +blk.10.attn_k_b.weight iq4_nl +blk.11.attn_k_b.weight iq4_nl +blk.12.attn_k_b.weight iq4_nl +blk.13.attn_k_b.weight iq4_nl +blk.14.attn_k_b.weight iq4_nl +blk.15.attn_k_b.weight iq4_nl +blk.16.attn_k_b.weight iq4_nl +blk.17.attn_k_b.weight iq4_nl +blk.18.attn_k_b.weight iq4_nl +blk.19.attn_k_b.weight iq4_nl +blk.20.attn_k_b.weight iq4_nl +blk.21.attn_k_b.weight iq4_nl +blk.22.attn_k_b.weight iq4_nl +blk.23.attn_k_b.weight iq4_nl +blk.24.attn_k_b.weight iq4_nl +blk.25.attn_k_b.weight iq4_nl +blk.26.attn_k_b.weight iq4_nl +blk.27.attn_k_b.weight iq4_nl +blk.28.attn_k_b.weight iq4_nl +blk.29.attn_k_b.weight iq4_nl +blk.30.attn_k_b.weight iq4_nl +blk.31.attn_k_b.weight iq4_nl +blk.32.attn_k_b.weight iq4_nl +blk.33.attn_k_b.weight iq4_nl +blk.34.attn_k_b.weight iq4_nl +blk.35.attn_k_b.weight iq4_nl +blk.36.attn_k_b.weight iq4_nl +blk.37.attn_k_b.weight iq4_nl +blk.38.attn_k_b.weight iq4_nl +blk.39.attn_k_b.weight iq4_nl +blk.40.attn_k_b.weight iq4_nl +blk.41.attn_k_b.weight iq4_nl +blk.42.attn_k_b.weight iq4_nl +blk.43.attn_k_b.weight iq4_nl +blk.44.attn_k_b.weight iq4_nl +blk.45.attn_k_b.weight iq4_nl +blk.46.attn_k_b.weight iq4_nl +blk.47.attn_k_b.weight iq4_nl +blk.48.attn_k_b.weight iq4_nl +blk.49.attn_k_b.weight iq4_nl +blk.50.attn_k_b.weight iq4_nl +blk.51.attn_k_b.weight iq4_nl +blk.52.attn_k_b.weight iq4_nl +blk.53.attn_k_b.weight iq4_nl +blk.54.attn_k_b.weight iq4_nl +blk.55.attn_k_b.weight iq4_nl +blk.56.attn_k_b.weight iq4_nl +blk.57.attn_k_b.weight iq4_nl +blk.58.attn_k_b.weight iq4_nl +blk.59.attn_k_b.weight iq4_nl +blk.60.attn_k_b.weight iq4_nl [IQ2_XS] iq2_xs output.weight q5_K token_embd.weight q2_K +blk.0.attn_k_b.weight iq4_nl blk.0.ffn_down.weight q2_K +blk.1.attn_k_b.weight iq4_nl blk.1.ffn_down.weight q2_K +blk.2.attn_k_b.weight iq4_nl blk.2.ffn_down.weight q2_K +blk.3.attn_k_b.weight iq4_nl blk.3.ffn_down_exps.weight q2_K blk.3.ffn_down_shexp.weight q2_K +blk.4.attn_k_b.weight iq4_nl blk.4.ffn_down_exps.weight q2_K blk.4.ffn_down_shexp.weight q2_K +blk.5.attn_k_b.weight iq4_nl +blk.6.attn_k_b.weight iq4_nl +blk.7.attn_k_b.weight iq4_nl +blk.8.attn_k_b.weight iq4_nl +blk.9.attn_k_b.weight iq4_nl +blk.10.attn_k_b.weight iq4_nl +blk.11.attn_k_b.weight iq4_nl +blk.12.attn_k_b.weight iq4_nl +blk.13.attn_k_b.weight iq4_nl +blk.14.attn_k_b.weight iq4_nl +blk.15.attn_k_b.weight iq4_nl +blk.16.attn_k_b.weight iq4_nl +blk.17.attn_k_b.weight iq4_nl +blk.18.attn_k_b.weight iq4_nl +blk.19.attn_k_b.weight iq4_nl +blk.20.attn_k_b.weight iq4_nl +blk.21.attn_k_b.weight iq4_nl +blk.22.attn_k_b.weight iq4_nl +blk.23.attn_k_b.weight iq4_nl +blk.24.attn_k_b.weight iq4_nl +blk.25.attn_k_b.weight iq4_nl +blk.26.attn_k_b.weight iq4_nl +blk.27.attn_k_b.weight iq4_nl +blk.28.attn_k_b.weight iq4_nl +blk.29.attn_k_b.weight iq4_nl +blk.30.attn_k_b.weight iq4_nl +blk.31.attn_k_b.weight iq4_nl +blk.32.attn_k_b.weight iq4_nl +blk.33.attn_k_b.weight iq4_nl +blk.34.attn_k_b.weight iq4_nl +blk.35.attn_k_b.weight iq4_nl +blk.36.attn_k_b.weight iq4_nl +blk.37.attn_k_b.weight iq4_nl +blk.38.attn_k_b.weight iq4_nl +blk.39.attn_k_b.weight iq4_nl +blk.40.attn_k_b.weight iq4_nl +blk.41.attn_k_b.weight iq4_nl +blk.42.attn_k_b.weight iq4_nl +blk.43.attn_k_b.weight iq4_nl +blk.44.attn_k_b.weight iq4_nl +blk.45.attn_k_b.weight iq4_nl +blk.46.attn_k_b.weight iq4_nl +blk.47.attn_k_b.weight iq4_nl +blk.48.attn_k_b.weight iq4_nl +blk.49.attn_k_b.weight iq4_nl +blk.50.attn_k_b.weight iq4_nl +blk.51.attn_k_b.weight iq4_nl +blk.52.attn_k_b.weight iq4_nl +blk.53.attn_k_b.weight iq4_nl +blk.54.attn_k_b.weight iq4_nl +blk.55.attn_k_b.weight iq4_nl +blk.56.attn_k_b.weight iq4_nl +blk.57.attn_k_b.weight iq4_nl +blk.58.attn_k_b.weight iq4_nl +blk.59.attn_k_b.weight iq4_nl +blk.60.attn_k_b.weight iq4_nl [Q2_K_S] q2_K output.weight q6_K +blk.0.attn_k_b.weight q4_0 blk.0.ffn_down.weight q4_K +blk.1.attn_k_b.weight q4_0 blk.1.ffn_down.weight q4_K +blk.2.attn_k_b.weight q4_0 blk.2.ffn_down.weight q4_K +blk.3.attn_k_b.weight q4_0 blk.3.ffn_down_exps.weight q4_K blk.3.ffn_down_shexp.weight q4_K +blk.4.attn_k_b.weight q4_0 blk.4.ffn_down_exps.weight q4_K blk.4.ffn_down_shexp.weight q4_K +blk.5.attn_k_b.weight q4_0 blk.5.ffn_down_exps.weight q4_K blk.5.ffn_down_shexp.weight q4_K +blk.6.attn_k_b.weight q4_0 blk.6.ffn_down_exps.weight q4_K blk.6.ffn_down_shexp.weight q4_K +blk.7.attn_k_b.weight q4_0 +blk.8.attn_k_b.weight q4_0 +blk.9.attn_k_b.weight q4_0 +blk.10.attn_k_b.weight q4_0 +blk.11.attn_k_b.weight q4_0 +blk.12.attn_k_b.weight q4_0 +blk.13.attn_k_b.weight q4_0 +blk.14.attn_k_b.weight q4_0 +blk.15.attn_k_b.weight q4_0 +blk.16.attn_k_b.weight q4_0 +blk.17.attn_k_b.weight q4_0 +blk.18.attn_k_b.weight q4_0 +blk.19.attn_k_b.weight q4_0 +blk.20.attn_k_b.weight q4_0 +blk.21.attn_k_b.weight q4_0 +blk.22.attn_k_b.weight q4_0 +blk.23.attn_k_b.weight q4_0 +blk.24.attn_k_b.weight q4_0 +blk.25.attn_k_b.weight q4_0 +blk.26.attn_k_b.weight q4_0 +blk.27.attn_k_b.weight q4_0 +blk.28.attn_k_b.weight q4_0 +blk.29.attn_k_b.weight q4_0 +blk.30.attn_k_b.weight q4_0 +blk.31.attn_k_b.weight q4_0 +blk.32.attn_k_b.weight q4_0 +blk.33.attn_k_b.weight q4_0 +blk.34.attn_k_b.weight q4_0 +blk.35.attn_k_b.weight q4_0 +blk.36.attn_k_b.weight q4_0 +blk.37.attn_k_b.weight q4_0 +blk.38.attn_k_b.weight q4_0 +blk.39.attn_k_b.weight q4_0 +blk.40.attn_k_b.weight q4_0 +blk.41.attn_k_b.weight q4_0 +blk.42.attn_k_b.weight q4_0 +blk.43.attn_k_b.weight q4_0 +blk.44.attn_k_b.weight q4_0 +blk.45.attn_k_b.weight q4_0 +blk.46.attn_k_b.weight q4_0 +blk.47.attn_k_b.weight q4_0 +blk.48.attn_k_b.weight q4_0 +blk.49.attn_k_b.weight q4_0 +blk.50.attn_k_b.weight q4_0 +blk.51.attn_k_b.weight q4_0 +blk.52.attn_k_b.weight q4_0 +blk.53.attn_k_b.weight q4_0 +blk.54.attn_k_b.weight q4_0 +blk.55.attn_k_b.weight q4_0 +blk.56.attn_k_b.weight q4_0 +blk.57.attn_k_b.weight q4_0 +blk.58.attn_k_b.weight q4_0 +blk.59.attn_k_b.weight q4_0 +blk.60.attn_k_b.weight q4_0 [IQ3_XS] iq3_s output.weight q6_K +blk.0.attn_k_b.weight iq4_nl +blk.1.attn_k_b.weight iq4_nl +blk.2.attn_k_b.weight iq4_nl +blk.3.attn_k_b.weight iq4_nl +blk.4.attn_k_b.weight iq4_nl +blk.5.attn_k_b.weight iq4_nl +blk.6.attn_k_b.weight iq4_nl +blk.7.attn_k_b.weight iq4_nl blk.7.ffn_gate_exps.weight iq3_xxs blk.7.ffn_gate_shexp.weight iq3_xxs blk.7.ffn_up_exps.weight iq3_xxs blk.7.ffn_up_shexp.weight iq3_xxs +blk.8.attn_k_b.weight iq4_nl blk.8.ffn_gate_exps.weight iq3_xxs blk.8.ffn_gate_shexp.weight iq3_xxs blk.8.ffn_up_exps.weight iq3_xxs blk.8.ffn_up_shexp.weight iq3_xxs +blk.9.attn_k_b.weight iq4_nl blk.9.ffn_gate_exps.weight iq3_xxs blk.9.ffn_gate_shexp.weight iq3_xxs blk.9.ffn_up_exps.weight iq3_xxs blk.9.ffn_up_shexp.weight iq3_xxs +blk.10.attn_k_b.weight iq4_nl blk.10.ffn_gate_exps.weight iq3_xxs blk.10.ffn_gate_shexp.weight iq3_xxs blk.10.ffn_up_exps.weight iq3_xxs blk.10.ffn_up_shexp.weight iq3_xxs +blk.11.attn_k_b.weight iq4_nl blk.11.ffn_gate_exps.weight iq3_xxs blk.11.ffn_gate_shexp.weight iq3_xxs blk.11.ffn_up_exps.weight iq3_xxs blk.11.ffn_up_shexp.weight iq3_xxs +blk.12.attn_k_b.weight iq4_nl blk.12.ffn_gate_exps.weight iq3_xxs blk.12.ffn_gate_shexp.weight iq3_xxs blk.12.ffn_up_exps.weight iq3_xxs blk.12.ffn_up_shexp.weight iq3_xxs +blk.13.attn_k_b.weight iq4_nl blk.13.ffn_gate_exps.weight iq3_xxs blk.13.ffn_gate_shexp.weight iq3_xxs blk.13.ffn_up_exps.weight iq3_xxs blk.13.ffn_up_shexp.weight iq3_xxs +blk.14.attn_k_b.weight iq4_nl blk.14.ffn_gate_exps.weight iq3_xxs blk.14.ffn_gate_shexp.weight iq3_xxs blk.14.ffn_up_exps.weight iq3_xxs blk.14.ffn_up_shexp.weight iq3_xxs +blk.15.attn_k_b.weight iq4_nl blk.15.ffn_gate_exps.weight iq3_xxs blk.15.ffn_gate_shexp.weight iq3_xxs blk.15.ffn_up_exps.weight iq3_xxs blk.15.ffn_up_shexp.weight iq3_xxs +blk.16.attn_k_b.weight iq4_nl blk.16.ffn_gate_exps.weight iq3_xxs blk.16.ffn_gate_shexp.weight iq3_xxs blk.16.ffn_up_exps.weight iq3_xxs blk.16.ffn_up_shexp.weight iq3_xxs +blk.17.attn_k_b.weight iq4_nl blk.17.ffn_gate_exps.weight iq3_xxs blk.17.ffn_gate_shexp.weight iq3_xxs blk.17.ffn_up_exps.weight iq3_xxs blk.17.ffn_up_shexp.weight iq3_xxs +blk.18.attn_k_b.weight iq4_nl blk.18.ffn_gate_exps.weight iq3_xxs blk.18.ffn_gate_shexp.weight iq3_xxs blk.18.ffn_up_exps.weight iq3_xxs blk.18.ffn_up_shexp.weight iq3_xxs +blk.19.attn_k_b.weight iq4_nl blk.19.ffn_gate_exps.weight iq3_xxs blk.19.ffn_gate_shexp.weight iq3_xxs blk.19.ffn_up_exps.weight iq3_xxs blk.19.ffn_up_shexp.weight iq3_xxs +blk.20.attn_k_b.weight iq4_nl blk.20.ffn_gate_exps.weight iq3_xxs blk.20.ffn_gate_shexp.weight iq3_xxs blk.20.ffn_up_exps.weight iq3_xxs blk.20.ffn_up_shexp.weight iq3_xxs +blk.21.attn_k_b.weight iq4_nl blk.21.ffn_gate_exps.weight iq3_xxs blk.21.ffn_gate_shexp.weight iq3_xxs blk.21.ffn_up_exps.weight iq3_xxs blk.21.ffn_up_shexp.weight iq3_xxs +blk.22.attn_k_b.weight iq4_nl blk.22.ffn_gate_exps.weight iq3_xxs blk.22.ffn_gate_shexp.weight iq3_xxs blk.22.ffn_up_exps.weight iq3_xxs blk.22.ffn_up_shexp.weight iq3_xxs +blk.23.attn_k_b.weight iq4_nl blk.23.ffn_gate_exps.weight iq3_xxs blk.23.ffn_gate_shexp.weight iq3_xxs blk.23.ffn_up_exps.weight iq3_xxs blk.23.ffn_up_shexp.weight iq3_xxs +blk.24.attn_k_b.weight iq4_nl blk.24.ffn_gate_exps.weight iq3_xxs blk.24.ffn_gate_shexp.weight iq3_xxs blk.24.ffn_up_exps.weight iq3_xxs blk.24.ffn_up_shexp.weight iq3_xxs +blk.25.attn_k_b.weight iq4_nl blk.25.ffn_gate_exps.weight iq3_xxs blk.25.ffn_gate_shexp.weight iq3_xxs blk.25.ffn_up_exps.weight iq3_xxs blk.25.ffn_up_shexp.weight iq3_xxs +blk.26.attn_k_b.weight iq4_nl blk.26.ffn_gate_exps.weight iq3_xxs blk.26.ffn_gate_shexp.weight iq3_xxs blk.26.ffn_up_exps.weight iq3_xxs blk.26.ffn_up_shexp.weight iq3_xxs +blk.27.attn_k_b.weight iq4_nl blk.27.ffn_gate_exps.weight iq3_xxs blk.27.ffn_gate_shexp.weight iq3_xxs blk.27.ffn_up_exps.weight iq3_xxs blk.27.ffn_up_shexp.weight iq3_xxs +blk.28.attn_k_b.weight iq4_nl blk.28.ffn_gate_exps.weight iq3_xxs blk.28.ffn_gate_shexp.weight iq3_xxs blk.28.ffn_up_exps.weight iq3_xxs blk.28.ffn_up_shexp.weight iq3_xxs +blk.29.attn_k_b.weight iq4_nl blk.29.ffn_gate_exps.weight iq3_xxs blk.29.ffn_gate_shexp.weight iq3_xxs blk.29.ffn_up_exps.weight iq3_xxs blk.29.ffn_up_shexp.weight iq3_xxs +blk.30.attn_k_b.weight iq4_nl blk.30.ffn_gate_exps.weight iq3_xxs blk.30.ffn_gate_shexp.weight iq3_xxs blk.30.ffn_up_exps.weight iq3_xxs blk.30.ffn_up_shexp.weight iq3_xxs +blk.31.attn_k_b.weight iq4_nl blk.31.ffn_gate_exps.weight iq3_xxs blk.31.ffn_gate_shexp.weight iq3_xxs blk.31.ffn_up_exps.weight iq3_xxs blk.31.ffn_up_shexp.weight iq3_xxs +blk.32.attn_k_b.weight iq4_nl blk.32.ffn_gate_exps.weight iq3_xxs blk.32.ffn_gate_shexp.weight iq3_xxs blk.32.ffn_up_exps.weight iq3_xxs blk.32.ffn_up_shexp.weight iq3_xxs +blk.33.attn_k_b.weight iq4_nl blk.33.ffn_gate_exps.weight iq3_xxs blk.33.ffn_gate_shexp.weight iq3_xxs blk.33.ffn_up_exps.weight iq3_xxs blk.33.ffn_up_shexp.weight iq3_xxs +blk.34.attn_k_b.weight iq4_nl blk.34.ffn_gate_exps.weight iq3_xxs blk.34.ffn_gate_shexp.weight iq3_xxs blk.34.ffn_up_exps.weight iq3_xxs blk.34.ffn_up_shexp.weight iq3_xxs +blk.35.attn_k_b.weight iq4_nl blk.35.ffn_gate_exps.weight iq3_xxs blk.35.ffn_gate_shexp.weight iq3_xxs blk.35.ffn_up_exps.weight iq3_xxs blk.35.ffn_up_shexp.weight iq3_xxs +blk.36.attn_k_b.weight iq4_nl blk.36.ffn_gate_exps.weight iq3_xxs blk.36.ffn_gate_shexp.weight iq3_xxs blk.36.ffn_up_exps.weight iq3_xxs blk.36.ffn_up_shexp.weight iq3_xxs +blk.37.attn_k_b.weight iq4_nl blk.37.ffn_gate_exps.weight iq3_xxs blk.37.ffn_gate_shexp.weight iq3_xxs blk.37.ffn_up_exps.weight iq3_xxs blk.37.ffn_up_shexp.weight iq3_xxs +blk.38.attn_k_b.weight iq4_nl blk.38.ffn_gate_exps.weight iq3_xxs blk.38.ffn_gate_shexp.weight iq3_xxs blk.38.ffn_up_exps.weight iq3_xxs blk.38.ffn_up_shexp.weight iq3_xxs +blk.39.attn_k_b.weight iq4_nl blk.39.ffn_gate_exps.weight iq3_xxs blk.39.ffn_gate_shexp.weight iq3_xxs blk.39.ffn_up_exps.weight iq3_xxs blk.39.ffn_up_shexp.weight iq3_xxs +blk.40.attn_k_b.weight iq4_nl blk.40.ffn_gate_exps.weight iq3_xxs blk.40.ffn_gate_shexp.weight iq3_xxs blk.40.ffn_up_exps.weight iq3_xxs blk.40.ffn_up_shexp.weight iq3_xxs +blk.41.attn_k_b.weight iq4_nl blk.41.ffn_gate_exps.weight iq3_xxs blk.41.ffn_gate_shexp.weight iq3_xxs blk.41.ffn_up_exps.weight iq3_xxs blk.41.ffn_up_shexp.weight iq3_xxs +blk.42.attn_k_b.weight iq4_nl blk.42.ffn_gate_exps.weight iq3_xxs blk.42.ffn_gate_shexp.weight iq3_xxs blk.42.ffn_up_exps.weight iq3_xxs blk.42.ffn_up_shexp.weight iq3_xxs +blk.43.attn_k_b.weight iq4_nl blk.43.ffn_gate_exps.weight iq3_xxs blk.43.ffn_gate_shexp.weight iq3_xxs blk.43.ffn_up_exps.weight iq3_xxs blk.43.ffn_up_shexp.weight iq3_xxs +blk.44.attn_k_b.weight iq4_nl blk.44.ffn_gate_exps.weight iq3_xxs blk.44.ffn_gate_shexp.weight iq3_xxs blk.44.ffn_up_exps.weight iq3_xxs blk.44.ffn_up_shexp.weight iq3_xxs +blk.45.attn_k_b.weight iq4_nl blk.45.ffn_gate_exps.weight iq3_xxs blk.45.ffn_gate_shexp.weight iq3_xxs blk.45.ffn_up_exps.weight iq3_xxs blk.45.ffn_up_shexp.weight iq3_xxs +blk.46.attn_k_b.weight iq4_nl blk.46.ffn_gate_exps.weight iq3_xxs blk.46.ffn_gate_shexp.weight iq3_xxs blk.46.ffn_up_exps.weight iq3_xxs blk.46.ffn_up_shexp.weight iq3_xxs +blk.47.attn_k_b.weight iq4_nl blk.47.ffn_gate_exps.weight iq3_xxs blk.47.ffn_gate_shexp.weight iq3_xxs blk.47.ffn_up_exps.weight iq3_xxs blk.47.ffn_up_shexp.weight iq3_xxs +blk.48.attn_k_b.weight iq4_nl blk.48.ffn_gate_exps.weight iq3_xxs blk.48.ffn_gate_shexp.weight iq3_xxs blk.48.ffn_up_exps.weight iq3_xxs blk.48.ffn_up_shexp.weight iq3_xxs +blk.49.attn_k_b.weight iq4_nl blk.49.ffn_gate_exps.weight iq3_xxs blk.49.ffn_gate_shexp.weight iq3_xxs blk.49.ffn_up_exps.weight iq3_xxs blk.49.ffn_up_shexp.weight iq3_xxs +blk.50.attn_k_b.weight iq4_nl blk.50.ffn_gate_exps.weight iq3_xxs blk.50.ffn_gate_shexp.weight iq3_xxs blk.50.ffn_up_exps.weight iq3_xxs blk.50.ffn_up_shexp.weight iq3_xxs +blk.51.attn_k_b.weight iq4_nl blk.51.ffn_gate_exps.weight iq3_xxs blk.51.ffn_gate_shexp.weight iq3_xxs blk.51.ffn_up_exps.weight iq3_xxs blk.51.ffn_up_shexp.weight iq3_xxs +blk.52.attn_k_b.weight iq4_nl blk.52.ffn_gate_exps.weight iq3_xxs blk.52.ffn_gate_shexp.weight iq3_xxs blk.52.ffn_up_exps.weight iq3_xxs blk.52.ffn_up_shexp.weight iq3_xxs +blk.53.attn_k_b.weight iq4_nl +blk.54.attn_k_b.weight iq4_nl +blk.55.attn_k_b.weight iq4_nl +blk.56.attn_k_b.weight iq4_nl +blk.57.attn_k_b.weight iq4_nl +blk.58.attn_k_b.weight iq4_nl +blk.59.attn_k_b.weight iq4_nl +blk.60.attn_k_b.weight iq4_nl [IQ3_XXS] iq3_xxs output.weight q5_K token_embd.weight iq3_s +blk.0.attn_k_b.weight iq4_nl blk.0.attn_output.weight iq3_s blk.0.ffn_down.weight q4_K +blk.1.attn_k_b.weight iq4_nl blk.1.attn_output.weight iq3_s blk.1.ffn_down.weight q4_K +blk.2.attn_k_b.weight iq4_nl blk.2.attn_output.weight iq3_s blk.2.ffn_down.weight q4_K +blk.3.attn_k_b.weight iq4_nl blk.3.attn_output.weight iq3_s blk.3.ffn_down_exps.weight q4_K blk.3.ffn_down_shexp.weight q4_K +blk.4.attn_k_b.weight iq4_nl blk.4.attn_output.weight iq3_s blk.4.ffn_down_exps.weight q4_K blk.4.ffn_down_shexp.weight q4_K +blk.5.attn_k_b.weight iq4_nl blk.5.attn_output.weight iq3_s blk.5.ffn_down_exps.weight q4_K blk.5.ffn_down_shexp.weight q4_K +blk.6.attn_k_b.weight iq4_nl blk.6.attn_output.weight iq3_s blk.6.ffn_down_exps.weight q4_K blk.6.ffn_down_shexp.weight q4_K +blk.7.attn_k_b.weight iq4_nl blk.7.attn_output.weight iq3_s blk.7.ffn_down_exps.weight q3_K blk.7.ffn_down_shexp.weight q3_K +blk.8.attn_k_b.weight iq4_nl blk.8.attn_output.weight iq3_s blk.8.ffn_down_exps.weight q3_K blk.8.ffn_down_shexp.weight q3_K +blk.9.attn_k_b.weight iq4_nl blk.9.attn_output.weight iq3_s blk.9.ffn_down_exps.weight q3_K blk.9.ffn_down_shexp.weight q3_K +blk.10.attn_k_b.weight iq4_nl blk.10.attn_output.weight iq3_s blk.10.ffn_down_exps.weight q3_K blk.10.ffn_down_shexp.weight q3_K +blk.11.attn_k_b.weight iq4_nl blk.11.attn_output.weight iq3_s blk.11.ffn_down_exps.weight q3_K blk.11.ffn_down_shexp.weight q3_K +blk.12.attn_k_b.weight iq4_nl blk.12.attn_output.weight iq3_s blk.12.ffn_down_exps.weight q3_K blk.12.ffn_down_shexp.weight q3_K +blk.13.attn_k_b.weight iq4_nl blk.13.attn_output.weight iq3_s blk.13.ffn_down_exps.weight q3_K blk.13.ffn_down_shexp.weight q3_K +blk.14.attn_k_b.weight iq4_nl blk.14.attn_output.weight iq3_s blk.14.ffn_down_exps.weight q3_K blk.14.ffn_down_shexp.weight q3_K +blk.15.attn_k_b.weight iq4_nl blk.15.attn_output.weight iq3_s blk.15.ffn_down_exps.weight q3_K blk.15.ffn_down_shexp.weight q3_K +blk.16.attn_k_b.weight iq4_nl blk.16.attn_output.weight iq3_s blk.16.ffn_down_exps.weight q3_K blk.16.ffn_down_shexp.weight q3_K +blk.17.attn_k_b.weight iq4_nl blk.17.attn_output.weight iq3_s blk.17.ffn_down_exps.weight q3_K blk.17.ffn_down_shexp.weight q3_K +blk.18.attn_k_b.weight iq4_nl blk.18.attn_output.weight iq3_s blk.18.ffn_down_exps.weight q3_K blk.18.ffn_down_shexp.weight q3_K +blk.19.attn_k_b.weight iq4_nl blk.19.attn_output.weight iq3_s blk.19.ffn_down_exps.weight q3_K blk.19.ffn_down_shexp.weight q3_K +blk.20.attn_k_b.weight iq4_nl blk.20.attn_output.weight iq3_s blk.20.ffn_down_exps.weight q3_K blk.20.ffn_down_shexp.weight q3_K +blk.21.attn_k_b.weight iq4_nl blk.21.attn_output.weight iq3_s blk.21.ffn_down_exps.weight q3_K blk.21.ffn_down_shexp.weight q3_K +blk.22.attn_k_b.weight iq4_nl blk.22.attn_output.weight iq3_s blk.22.ffn_down_exps.weight q3_K blk.22.ffn_down_shexp.weight q3_K +blk.23.attn_k_b.weight iq4_nl blk.23.attn_output.weight iq3_s blk.23.ffn_down_exps.weight q3_K blk.23.ffn_down_shexp.weight q3_K +blk.24.attn_k_b.weight iq4_nl blk.24.attn_output.weight iq3_s blk.24.ffn_down_exps.weight q3_K blk.24.ffn_down_shexp.weight q3_K +blk.25.attn_k_b.weight iq4_nl blk.25.attn_output.weight iq3_s blk.25.ffn_down_exps.weight q3_K blk.25.ffn_down_shexp.weight q3_K +blk.26.attn_k_b.weight iq4_nl blk.26.attn_output.weight iq3_s blk.26.ffn_down_exps.weight q3_K blk.26.ffn_down_shexp.weight q3_K +blk.27.attn_k_b.weight iq4_nl blk.27.attn_output.weight iq3_s blk.27.ffn_down_exps.weight q3_K blk.27.ffn_down_shexp.weight q3_K +blk.28.attn_k_b.weight iq4_nl blk.28.attn_output.weight iq3_s blk.28.ffn_down_exps.weight q3_K blk.28.ffn_down_shexp.weight q3_K +blk.29.attn_k_b.weight iq4_nl blk.29.attn_output.weight iq3_s blk.29.ffn_down_exps.weight q3_K blk.29.ffn_down_shexp.weight q3_K +blk.30.attn_k_b.weight iq4_nl blk.30.attn_output.weight iq3_s blk.30.ffn_down_exps.weight q3_K blk.30.ffn_down_shexp.weight q3_K +blk.31.attn_k_b.weight iq4_nl blk.31.attn_output.weight iq3_s blk.31.ffn_down_exps.weight q3_K blk.31.ffn_down_shexp.weight q3_K +blk.32.attn_k_b.weight iq4_nl blk.32.attn_output.weight iq3_s blk.32.ffn_down_exps.weight q3_K blk.32.ffn_down_shexp.weight q3_K +blk.33.attn_k_b.weight iq4_nl blk.33.attn_output.weight iq3_s blk.33.ffn_down_exps.weight q3_K blk.33.ffn_down_shexp.weight q3_K +blk.34.attn_k_b.weight iq4_nl blk.34.attn_output.weight iq3_s blk.34.ffn_down_exps.weight q3_K blk.34.ffn_down_shexp.weight q3_K +blk.35.attn_k_b.weight iq4_nl blk.35.attn_output.weight iq3_s blk.35.ffn_down_exps.weight q3_K blk.35.ffn_down_shexp.weight q3_K +blk.36.attn_k_b.weight iq4_nl blk.36.attn_output.weight iq3_s blk.36.ffn_down_exps.weight q3_K blk.36.ffn_down_shexp.weight q3_K +blk.37.attn_k_b.weight iq4_nl blk.37.attn_output.weight iq3_s blk.37.ffn_down_exps.weight q3_K blk.37.ffn_down_shexp.weight q3_K +blk.38.attn_k_b.weight iq4_nl blk.38.attn_output.weight iq3_s blk.38.ffn_down_exps.weight q3_K blk.38.ffn_down_shexp.weight q3_K +blk.39.attn_k_b.weight iq4_nl blk.39.attn_output.weight iq3_s blk.39.ffn_down_exps.weight q3_K blk.39.ffn_down_shexp.weight q3_K +blk.40.attn_k_b.weight iq4_nl blk.40.attn_output.weight iq3_s blk.40.ffn_down_exps.weight q3_K blk.40.ffn_down_shexp.weight q3_K +blk.41.attn_k_b.weight iq4_nl blk.41.attn_output.weight iq3_s blk.41.ffn_down_exps.weight q3_K blk.41.ffn_down_shexp.weight q3_K +blk.42.attn_k_b.weight iq4_nl blk.42.attn_output.weight iq3_s blk.42.ffn_down_exps.weight q3_K blk.42.ffn_down_shexp.weight q3_K +blk.43.attn_k_b.weight iq4_nl blk.43.attn_output.weight iq3_s blk.43.ffn_down_exps.weight q3_K blk.43.ffn_down_shexp.weight q3_K +blk.44.attn_k_b.weight iq4_nl blk.44.attn_output.weight iq3_s blk.44.ffn_down_exps.weight q3_K blk.44.ffn_down_shexp.weight q3_K +blk.45.attn_k_b.weight iq4_nl blk.45.attn_output.weight iq3_s blk.45.ffn_down_exps.weight q3_K blk.45.ffn_down_shexp.weight q3_K +blk.46.attn_k_b.weight iq4_nl blk.46.attn_output.weight iq3_s blk.46.ffn_down_exps.weight q3_K blk.46.ffn_down_shexp.weight q3_K +blk.47.attn_k_b.weight iq4_nl blk.47.attn_output.weight iq3_s blk.47.ffn_down_exps.weight q3_K blk.47.ffn_down_shexp.weight q3_K +blk.48.attn_k_b.weight iq4_nl blk.48.attn_output.weight iq3_s blk.48.ffn_down_exps.weight q3_K blk.48.ffn_down_shexp.weight q3_K +blk.49.attn_k_b.weight iq4_nl blk.49.attn_output.weight iq3_s blk.49.ffn_down_exps.weight q3_K blk.49.ffn_down_shexp.weight q3_K +blk.50.attn_k_b.weight iq4_nl blk.50.attn_output.weight iq3_s blk.50.ffn_down_exps.weight q3_K blk.50.ffn_down_shexp.weight q3_K +blk.51.attn_k_b.weight iq4_nl blk.51.attn_output.weight iq3_s blk.51.ffn_down_exps.weight q3_K blk.51.ffn_down_shexp.weight q3_K +blk.52.attn_k_b.weight iq4_nl blk.52.attn_output.weight iq3_s blk.52.ffn_down_exps.weight q3_K blk.52.ffn_down_shexp.weight q3_K +blk.53.attn_k_b.weight iq4_nl blk.53.attn_output.weight iq3_s blk.53.ffn_down_exps.weight q3_K blk.53.ffn_down_shexp.weight q3_K +blk.54.attn_k_b.weight iq4_nl blk.54.attn_output.weight iq3_s blk.54.ffn_down_exps.weight q3_K blk.54.ffn_down_shexp.weight q3_K +blk.55.attn_k_b.weight iq4_nl blk.55.attn_output.weight iq3_s blk.55.ffn_down_exps.weight q3_K blk.55.ffn_down_shexp.weight q3_K +blk.56.attn_k_b.weight iq4_nl blk.56.attn_output.weight iq3_s blk.56.ffn_down_exps.weight q3_K blk.56.ffn_down_shexp.weight q3_K +blk.57.attn_k_b.weight iq4_nl blk.57.attn_output.weight iq3_s blk.57.ffn_down_exps.weight q3_K blk.57.ffn_down_shexp.weight q3_K +blk.58.attn_k_b.weight iq4_nl blk.58.attn_output.weight iq3_s blk.58.ffn_down_exps.weight q3_K blk.58.ffn_down_shexp.weight q3_K +blk.59.attn_k_b.weight iq4_nl blk.59.attn_output.weight iq3_s blk.59.ffn_down_exps.weight q3_K blk.59.ffn_down_shexp.weight q3_K +blk.60.attn_k_b.weight iq4_nl blk.60.attn_output.weight iq3_s blk.60.ffn_down_exps.weight q3_K blk.60.ffn_down_shexp.weight q3_K @@ -1122,73 +1974,134 @@ blk.60.ffn_down_shexp.weight q3_K [IQ1_S] iq1_s output.weight q5_K token_embd.weight q2_K +blk.0.attn_k_b.weight iq4_nl blk.0.attn_output.weight iq2_xxs blk.0.ffn_down.weight q2_K +blk.1.attn_k_b.weight iq4_nl blk.1.attn_output.weight iq2_xxs blk.1.ffn_down.weight q2_K +blk.2.attn_k_b.weight iq4_nl blk.2.attn_output.weight iq2_xxs blk.2.ffn_down.weight q2_K +blk.3.attn_k_b.weight iq4_nl blk.3.attn_output.weight iq2_xxs blk.3.ffn_down_exps.weight q2_K blk.3.ffn_down_shexp.weight q2_K +blk.4.attn_k_b.weight iq4_nl blk.4.attn_output.weight iq2_xxs blk.4.ffn_down_exps.weight q2_K blk.4.ffn_down_shexp.weight q2_K +blk.5.attn_k_b.weight iq4_nl blk.5.attn_output.weight iq2_xxs +blk.6.attn_k_b.weight iq4_nl blk.6.attn_output.weight iq2_xxs +blk.7.attn_k_b.weight iq4_nl blk.7.attn_output.weight iq2_xxs +blk.8.attn_k_b.weight iq4_nl blk.8.attn_output.weight iq2_xxs +blk.9.attn_k_b.weight iq4_nl blk.9.attn_output.weight iq2_xxs +blk.10.attn_k_b.weight iq4_nl blk.10.attn_output.weight iq2_xxs +blk.11.attn_k_b.weight iq4_nl blk.11.attn_output.weight iq2_xxs +blk.12.attn_k_b.weight iq4_nl blk.12.attn_output.weight iq2_xxs +blk.13.attn_k_b.weight iq4_nl blk.13.attn_output.weight iq2_xxs +blk.14.attn_k_b.weight iq4_nl blk.14.attn_output.weight iq2_xxs +blk.15.attn_k_b.weight iq4_nl blk.15.attn_output.weight iq2_xxs +blk.16.attn_k_b.weight iq4_nl blk.16.attn_output.weight iq2_xxs +blk.17.attn_k_b.weight iq4_nl blk.17.attn_output.weight iq2_xxs +blk.18.attn_k_b.weight iq4_nl blk.18.attn_output.weight iq2_xxs +blk.19.attn_k_b.weight iq4_nl blk.19.attn_output.weight iq2_xxs +blk.20.attn_k_b.weight iq4_nl blk.20.attn_output.weight iq2_xxs +blk.21.attn_k_b.weight iq4_nl blk.21.attn_output.weight iq2_xxs +blk.22.attn_k_b.weight iq4_nl blk.22.attn_output.weight iq2_xxs +blk.23.attn_k_b.weight iq4_nl blk.23.attn_output.weight iq2_xxs +blk.24.attn_k_b.weight iq4_nl blk.24.attn_output.weight iq2_xxs +blk.25.attn_k_b.weight iq4_nl blk.25.attn_output.weight iq2_xxs +blk.26.attn_k_b.weight iq4_nl blk.26.attn_output.weight iq2_xxs +blk.27.attn_k_b.weight iq4_nl blk.27.attn_output.weight iq2_xxs +blk.28.attn_k_b.weight iq4_nl blk.28.attn_output.weight iq2_xxs +blk.29.attn_k_b.weight iq4_nl blk.29.attn_output.weight iq2_xxs +blk.30.attn_k_b.weight iq4_nl blk.30.attn_output.weight iq2_xxs +blk.31.attn_k_b.weight iq4_nl blk.31.attn_output.weight iq2_xxs +blk.32.attn_k_b.weight iq4_nl blk.32.attn_output.weight iq2_xxs +blk.33.attn_k_b.weight iq4_nl blk.33.attn_output.weight iq2_xxs +blk.34.attn_k_b.weight iq4_nl blk.34.attn_output.weight iq2_xxs +blk.35.attn_k_b.weight iq4_nl blk.35.attn_output.weight iq2_xxs +blk.36.attn_k_b.weight iq4_nl blk.36.attn_output.weight iq2_xxs +blk.37.attn_k_b.weight iq4_nl blk.37.attn_output.weight iq2_xxs +blk.38.attn_k_b.weight iq4_nl blk.38.attn_output.weight iq2_xxs +blk.39.attn_k_b.weight iq4_nl blk.39.attn_output.weight iq2_xxs +blk.40.attn_k_b.weight iq4_nl blk.40.attn_output.weight iq2_xxs +blk.41.attn_k_b.weight iq4_nl blk.41.attn_output.weight iq2_xxs +blk.42.attn_k_b.weight iq4_nl blk.42.attn_output.weight iq2_xxs +blk.43.attn_k_b.weight iq4_nl blk.43.attn_output.weight iq2_xxs +blk.44.attn_k_b.weight iq4_nl blk.44.attn_output.weight iq2_xxs +blk.45.attn_k_b.weight iq4_nl blk.45.attn_output.weight iq2_xxs +blk.46.attn_k_b.weight iq4_nl blk.46.attn_output.weight iq2_xxs +blk.47.attn_k_b.weight iq4_nl blk.47.attn_output.weight iq2_xxs +blk.48.attn_k_b.weight iq4_nl blk.48.attn_output.weight iq2_xxs +blk.49.attn_k_b.weight iq4_nl blk.49.attn_output.weight iq2_xxs +blk.50.attn_k_b.weight iq4_nl blk.50.attn_output.weight iq2_xxs +blk.51.attn_k_b.weight iq4_nl blk.51.attn_output.weight iq2_xxs +blk.52.attn_k_b.weight iq4_nl blk.52.attn_output.weight iq2_xxs +blk.53.attn_k_b.weight iq4_nl blk.53.attn_output.weight iq2_xxs +blk.54.attn_k_b.weight iq4_nl blk.54.attn_output.weight iq2_xxs +blk.55.attn_k_b.weight iq4_nl blk.55.attn_output.weight iq2_xxs +blk.56.attn_k_b.weight iq4_nl blk.56.attn_output.weight iq2_xxs +blk.57.attn_k_b.weight iq4_nl blk.57.attn_output.weight iq2_xxs +blk.58.attn_k_b.weight iq4_nl blk.58.attn_output.weight iq2_xxs +blk.59.attn_k_b.weight iq4_nl blk.59.attn_output.weight iq2_xxs +blk.60.attn_k_b.weight iq4_nl blk.60.attn_output.weight iq2_xxs [IQ4_NL] iq4_nl @@ -1207,322 +2120,809 @@ blk.6.ffn_down_shexp.weight q5_K [IQ3_S] iq3_s output.weight q6_K +blk.0.attn_k_b.weight iq4_nl +blk.1.attn_k_b.weight iq4_nl +blk.2.attn_k_b.weight iq4_nl +blk.3.attn_k_b.weight iq4_nl +blk.4.attn_k_b.weight iq4_nl +blk.5.attn_k_b.weight iq4_nl +blk.6.attn_k_b.weight iq4_nl +blk.7.attn_k_b.weight iq4_nl +blk.8.attn_k_b.weight iq4_nl +blk.9.attn_k_b.weight iq4_nl +blk.10.attn_k_b.weight iq4_nl +blk.11.attn_k_b.weight iq4_nl +blk.12.attn_k_b.weight iq4_nl +blk.13.attn_k_b.weight iq4_nl +blk.14.attn_k_b.weight iq4_nl +blk.15.attn_k_b.weight iq4_nl +blk.16.attn_k_b.weight iq4_nl +blk.17.attn_k_b.weight iq4_nl +blk.18.attn_k_b.weight iq4_nl +blk.19.attn_k_b.weight iq4_nl +blk.20.attn_k_b.weight iq4_nl +blk.21.attn_k_b.weight iq4_nl +blk.22.attn_k_b.weight iq4_nl +blk.23.attn_k_b.weight iq4_nl +blk.24.attn_k_b.weight iq4_nl +blk.25.attn_k_b.weight iq4_nl +blk.26.attn_k_b.weight iq4_nl +blk.27.attn_k_b.weight iq4_nl +blk.28.attn_k_b.weight iq4_nl +blk.29.attn_k_b.weight iq4_nl +blk.30.attn_k_b.weight iq4_nl +blk.31.attn_k_b.weight iq4_nl +blk.32.attn_k_b.weight iq4_nl +blk.33.attn_k_b.weight iq4_nl +blk.34.attn_k_b.weight iq4_nl +blk.35.attn_k_b.weight iq4_nl +blk.36.attn_k_b.weight iq4_nl +blk.37.attn_k_b.weight iq4_nl +blk.38.attn_k_b.weight iq4_nl +blk.39.attn_k_b.weight iq4_nl +blk.40.attn_k_b.weight iq4_nl +blk.41.attn_k_b.weight iq4_nl +blk.42.attn_k_b.weight iq4_nl +blk.43.attn_k_b.weight iq4_nl +blk.44.attn_k_b.weight iq4_nl +blk.45.attn_k_b.weight iq4_nl +blk.46.attn_k_b.weight iq4_nl +blk.47.attn_k_b.weight iq4_nl +blk.48.attn_k_b.weight iq4_nl +blk.49.attn_k_b.weight iq4_nl +blk.50.attn_k_b.weight iq4_nl +blk.51.attn_k_b.weight iq4_nl +blk.52.attn_k_b.weight iq4_nl +blk.53.attn_k_b.weight iq4_nl +blk.54.attn_k_b.weight iq4_nl +blk.55.attn_k_b.weight iq4_nl +blk.56.attn_k_b.weight iq4_nl +blk.57.attn_k_b.weight iq4_nl +blk.58.attn_k_b.weight iq4_nl +blk.59.attn_k_b.weight iq4_nl +blk.60.attn_k_b.weight iq4_nl [IQ3_M] iq3_s output.weight q6_K +blk.0.attn_k_b.weight iq4_nl blk.0.attn_output.weight q4_K blk.0.ffn_down.weight q4_K +blk.1.attn_k_b.weight iq4_nl blk.1.attn_output.weight q4_K blk.1.ffn_down.weight q4_K +blk.2.attn_k_b.weight iq4_nl blk.2.attn_output.weight q4_K blk.2.ffn_down.weight q4_K +blk.3.attn_k_b.weight iq4_nl blk.3.attn_output.weight q4_K blk.3.ffn_down_exps.weight q4_K blk.3.ffn_down_shexp.weight q4_K +blk.4.attn_k_b.weight iq4_nl blk.4.attn_output.weight q4_K blk.4.ffn_down_exps.weight q4_K blk.4.ffn_down_shexp.weight q4_K +blk.5.attn_k_b.weight iq4_nl blk.5.attn_output.weight q4_K blk.5.ffn_down_exps.weight q4_K blk.5.ffn_down_shexp.weight q4_K +blk.6.attn_k_b.weight iq4_nl blk.6.attn_output.weight q4_K blk.6.ffn_down_exps.weight q4_K blk.6.ffn_down_shexp.weight q4_K +blk.7.attn_k_b.weight iq4_nl blk.7.attn_output.weight q4_K +blk.8.attn_k_b.weight iq4_nl blk.8.attn_output.weight q4_K +blk.9.attn_k_b.weight iq4_nl blk.9.attn_output.weight q4_K +blk.10.attn_k_b.weight iq4_nl blk.10.attn_output.weight q4_K +blk.11.attn_k_b.weight iq4_nl blk.11.attn_output.weight q4_K +blk.12.attn_k_b.weight iq4_nl blk.12.attn_output.weight q4_K +blk.13.attn_k_b.weight iq4_nl blk.13.attn_output.weight q4_K +blk.14.attn_k_b.weight iq4_nl blk.14.attn_output.weight q4_K +blk.15.attn_k_b.weight iq4_nl blk.15.attn_output.weight q4_K +blk.16.attn_k_b.weight iq4_nl blk.16.attn_output.weight q4_K +blk.17.attn_k_b.weight iq4_nl blk.17.attn_output.weight q4_K +blk.18.attn_k_b.weight iq4_nl blk.18.attn_output.weight q4_K +blk.19.attn_k_b.weight iq4_nl blk.19.attn_output.weight q4_K +blk.20.attn_k_b.weight iq4_nl blk.20.attn_output.weight q4_K +blk.21.attn_k_b.weight iq4_nl blk.21.attn_output.weight q4_K +blk.22.attn_k_b.weight iq4_nl blk.22.attn_output.weight q4_K +blk.23.attn_k_b.weight iq4_nl blk.23.attn_output.weight q4_K +blk.24.attn_k_b.weight iq4_nl blk.24.attn_output.weight q4_K +blk.25.attn_k_b.weight iq4_nl blk.25.attn_output.weight q4_K +blk.26.attn_k_b.weight iq4_nl blk.26.attn_output.weight q4_K +blk.27.attn_k_b.weight iq4_nl blk.27.attn_output.weight q4_K +blk.28.attn_k_b.weight iq4_nl blk.28.attn_output.weight q4_K +blk.29.attn_k_b.weight iq4_nl blk.29.attn_output.weight q4_K +blk.30.attn_k_b.weight iq4_nl blk.30.attn_output.weight q4_K +blk.31.attn_k_b.weight iq4_nl blk.31.attn_output.weight q4_K +blk.32.attn_k_b.weight iq4_nl blk.32.attn_output.weight q4_K +blk.33.attn_k_b.weight iq4_nl blk.33.attn_output.weight q4_K +blk.34.attn_k_b.weight iq4_nl blk.34.attn_output.weight q4_K +blk.35.attn_k_b.weight iq4_nl blk.35.attn_output.weight q4_K +blk.36.attn_k_b.weight iq4_nl blk.36.attn_output.weight q4_K +blk.37.attn_k_b.weight iq4_nl blk.37.attn_output.weight q4_K +blk.38.attn_k_b.weight iq4_nl blk.38.attn_output.weight q4_K +blk.39.attn_k_b.weight iq4_nl blk.39.attn_output.weight q4_K +blk.40.attn_k_b.weight iq4_nl blk.40.attn_output.weight q4_K +blk.41.attn_k_b.weight iq4_nl blk.41.attn_output.weight q4_K +blk.42.attn_k_b.weight iq4_nl blk.42.attn_output.weight q4_K +blk.43.attn_k_b.weight iq4_nl blk.43.attn_output.weight q4_K +blk.44.attn_k_b.weight iq4_nl blk.44.attn_output.weight q4_K +blk.45.attn_k_b.weight iq4_nl blk.45.attn_output.weight q4_K +blk.46.attn_k_b.weight iq4_nl blk.46.attn_output.weight q4_K +blk.47.attn_k_b.weight iq4_nl blk.47.attn_output.weight q4_K +blk.48.attn_k_b.weight iq4_nl blk.48.attn_output.weight q4_K +blk.49.attn_k_b.weight iq4_nl blk.49.attn_output.weight q4_K +blk.50.attn_k_b.weight iq4_nl blk.50.attn_output.weight q4_K +blk.51.attn_k_b.weight iq4_nl blk.51.attn_output.weight q4_K +blk.52.attn_k_b.weight iq4_nl blk.52.attn_output.weight q4_K +blk.53.attn_k_b.weight iq4_nl blk.53.attn_output.weight q4_K +blk.54.attn_k_b.weight iq4_nl blk.54.attn_output.weight q4_K +blk.55.attn_k_b.weight iq4_nl blk.55.attn_output.weight q4_K +blk.56.attn_k_b.weight iq4_nl blk.56.attn_output.weight q4_K +blk.57.attn_k_b.weight iq4_nl blk.57.attn_output.weight q4_K +blk.58.attn_k_b.weight iq4_nl blk.58.attn_output.weight q4_K +blk.59.attn_k_b.weight iq4_nl blk.59.attn_output.weight q4_K +blk.60.attn_k_b.weight iq4_nl blk.60.attn_output.weight q4_K [IQ2_S] iq2_xs output.weight q5_K token_embd.weight iq3_s +blk.0.attn_k_b.weight iq4_nl blk.0.attn_output.weight iq3_s blk.0.ffn_down.weight iq3_s +blk.1.attn_k_b.weight iq4_nl blk.1.attn_output.weight iq3_s blk.1.ffn_down.weight iq3_s +blk.2.attn_k_b.weight iq4_nl blk.2.attn_output.weight iq3_s blk.2.ffn_down.weight iq3_s +blk.3.attn_k_b.weight iq4_nl blk.3.attn_output.weight iq3_s blk.3.ffn_down_exps.weight iq3_s blk.3.ffn_down_shexp.weight iq3_s +blk.4.attn_k_b.weight iq4_nl blk.4.attn_output.weight iq3_s blk.4.ffn_down_exps.weight iq3_s blk.4.ffn_down_shexp.weight iq3_s +blk.5.attn_k_b.weight iq4_nl blk.5.attn_output.weight iq3_s +blk.6.attn_k_b.weight iq4_nl blk.6.attn_output.weight iq3_s +blk.7.attn_k_b.weight iq4_nl blk.7.attn_output.weight iq3_s +blk.8.attn_k_b.weight iq4_nl blk.8.attn_output.weight iq3_s +blk.9.attn_k_b.weight iq4_nl blk.9.attn_output.weight iq3_s +blk.10.attn_k_b.weight iq4_nl blk.10.attn_output.weight iq3_s +blk.11.attn_k_b.weight iq4_nl blk.11.attn_output.weight iq3_s +blk.12.attn_k_b.weight iq4_nl blk.12.attn_output.weight iq3_s +blk.13.attn_k_b.weight iq4_nl blk.13.attn_output.weight iq3_s +blk.14.attn_k_b.weight iq4_nl blk.14.attn_output.weight iq3_s +blk.15.attn_k_b.weight iq4_nl blk.15.attn_output.weight iq3_s +blk.16.attn_k_b.weight iq4_nl blk.16.attn_output.weight iq3_s +blk.17.attn_k_b.weight iq4_nl blk.17.attn_output.weight iq3_s +blk.18.attn_k_b.weight iq4_nl blk.18.attn_output.weight iq3_s +blk.19.attn_k_b.weight iq4_nl blk.19.attn_output.weight iq3_s +blk.20.attn_k_b.weight iq4_nl blk.20.attn_output.weight iq3_s +blk.21.attn_k_b.weight iq4_nl blk.21.attn_output.weight iq3_s +blk.22.attn_k_b.weight iq4_nl blk.22.attn_output.weight iq3_s +blk.23.attn_k_b.weight iq4_nl blk.23.attn_output.weight iq3_s +blk.24.attn_k_b.weight iq4_nl blk.24.attn_output.weight iq3_s +blk.25.attn_k_b.weight iq4_nl blk.25.attn_output.weight iq3_s +blk.26.attn_k_b.weight iq4_nl blk.26.attn_output.weight iq3_s +blk.27.attn_k_b.weight iq4_nl blk.27.attn_output.weight iq3_s +blk.28.attn_k_b.weight iq4_nl blk.28.attn_output.weight iq3_s +blk.29.attn_k_b.weight iq4_nl blk.29.attn_output.weight iq3_s +blk.30.attn_k_b.weight iq4_nl blk.30.attn_output.weight iq3_s +blk.31.attn_k_b.weight iq4_nl blk.31.attn_output.weight iq3_s +blk.32.attn_k_b.weight iq4_nl blk.32.attn_output.weight iq3_s +blk.33.attn_k_b.weight iq4_nl blk.33.attn_output.weight iq3_s +blk.34.attn_k_b.weight iq4_nl blk.34.attn_output.weight iq3_s +blk.35.attn_k_b.weight iq4_nl blk.35.attn_output.weight iq3_s +blk.36.attn_k_b.weight iq4_nl blk.36.attn_output.weight iq3_s +blk.37.attn_k_b.weight iq4_nl blk.37.attn_output.weight iq3_s +blk.38.attn_k_b.weight iq4_nl blk.38.attn_output.weight iq3_s +blk.39.attn_k_b.weight iq4_nl blk.39.attn_output.weight iq3_s +blk.40.attn_k_b.weight iq4_nl blk.40.attn_output.weight iq3_s +blk.41.attn_k_b.weight iq4_nl blk.41.attn_output.weight iq3_s +blk.42.attn_k_b.weight iq4_nl blk.42.attn_output.weight iq3_s +blk.43.attn_k_b.weight iq4_nl blk.43.attn_output.weight iq3_s +blk.44.attn_k_b.weight iq4_nl blk.44.attn_output.weight iq3_s +blk.45.attn_k_b.weight iq4_nl blk.45.attn_output.weight iq3_s +blk.46.attn_k_b.weight iq4_nl blk.46.attn_output.weight iq3_s +blk.47.attn_k_b.weight iq4_nl blk.47.attn_output.weight iq3_s +blk.48.attn_k_b.weight iq4_nl blk.48.attn_output.weight iq3_s +blk.49.attn_k_b.weight iq4_nl blk.49.attn_output.weight iq3_s +blk.50.attn_k_b.weight iq4_nl blk.50.attn_output.weight iq3_s +blk.51.attn_k_b.weight iq4_nl blk.51.attn_output.weight iq3_s +blk.52.attn_k_b.weight iq4_nl blk.52.attn_output.weight iq3_s +blk.53.attn_k_b.weight iq4_nl blk.53.attn_output.weight iq3_s +blk.54.attn_k_b.weight iq4_nl blk.54.attn_output.weight iq3_s +blk.55.attn_k_b.weight iq4_nl blk.55.attn_output.weight iq3_s +blk.56.attn_k_b.weight iq4_nl blk.56.attn_output.weight iq3_s +blk.57.attn_k_b.weight iq4_nl blk.57.attn_output.weight iq3_s +blk.58.attn_k_b.weight iq4_nl blk.58.attn_output.weight iq3_s +blk.59.attn_k_b.weight iq4_nl blk.59.attn_output.weight iq3_s +blk.60.attn_k_b.weight iq4_nl blk.60.attn_output.weight iq3_s [IQ2_M] iq2_s output.weight q5_K token_embd.weight iq3_s +blk.0.attn_k_b.weight iq4_nl blk.0.attn_output.weight iq3_s blk.0.ffn_down.weight iq3_s +blk.1.attn_k_b.weight iq4_nl blk.1.attn_output.weight iq3_s blk.1.ffn_down.weight iq3_s +blk.2.attn_k_b.weight iq4_nl blk.2.attn_output.weight iq3_s blk.2.ffn_down.weight iq3_s +blk.3.attn_k_b.weight iq4_nl blk.3.attn_output.weight iq3_s blk.3.ffn_down_exps.weight iq3_s blk.3.ffn_down_shexp.weight iq3_s +blk.4.attn_k_b.weight iq4_nl blk.4.attn_output.weight iq3_s blk.4.ffn_down_exps.weight iq3_s blk.4.ffn_down_shexp.weight iq3_s +blk.5.attn_k_b.weight iq4_nl blk.5.attn_output.weight iq3_s +blk.6.attn_k_b.weight iq4_nl blk.6.attn_output.weight iq3_s +blk.7.attn_k_b.weight iq4_nl blk.7.attn_output.weight iq3_s +blk.8.attn_k_b.weight iq4_nl blk.8.attn_output.weight iq3_s +blk.9.attn_k_b.weight iq4_nl blk.9.attn_output.weight iq3_s +blk.10.attn_k_b.weight iq4_nl blk.10.attn_output.weight iq3_s +blk.11.attn_k_b.weight iq4_nl blk.11.attn_output.weight iq3_s +blk.12.attn_k_b.weight iq4_nl blk.12.attn_output.weight iq3_s +blk.13.attn_k_b.weight iq4_nl blk.13.attn_output.weight iq3_s +blk.14.attn_k_b.weight iq4_nl blk.14.attn_output.weight iq3_s +blk.15.attn_k_b.weight iq4_nl blk.15.attn_output.weight iq3_s +blk.16.attn_k_b.weight iq4_nl blk.16.attn_output.weight iq3_s +blk.17.attn_k_b.weight iq4_nl blk.17.attn_output.weight iq3_s +blk.18.attn_k_b.weight iq4_nl blk.18.attn_output.weight iq3_s +blk.19.attn_k_b.weight iq4_nl blk.19.attn_output.weight iq3_s +blk.20.attn_k_b.weight iq4_nl blk.20.attn_output.weight iq3_s +blk.21.attn_k_b.weight iq4_nl blk.21.attn_output.weight iq3_s +blk.22.attn_k_b.weight iq4_nl blk.22.attn_output.weight iq3_s +blk.23.attn_k_b.weight iq4_nl blk.23.attn_output.weight iq3_s +blk.24.attn_k_b.weight iq4_nl blk.24.attn_output.weight iq3_s +blk.25.attn_k_b.weight iq4_nl blk.25.attn_output.weight iq3_s +blk.26.attn_k_b.weight iq4_nl blk.26.attn_output.weight iq3_s +blk.27.attn_k_b.weight iq4_nl blk.27.attn_output.weight iq3_s +blk.28.attn_k_b.weight iq4_nl blk.28.attn_output.weight iq3_s +blk.29.attn_k_b.weight iq4_nl blk.29.attn_output.weight iq3_s +blk.30.attn_k_b.weight iq4_nl blk.30.attn_output.weight iq3_s +blk.31.attn_k_b.weight iq4_nl blk.31.attn_output.weight iq3_s +blk.32.attn_k_b.weight iq4_nl blk.32.attn_output.weight iq3_s +blk.33.attn_k_b.weight iq4_nl blk.33.attn_output.weight iq3_s +blk.34.attn_k_b.weight iq4_nl blk.34.attn_output.weight iq3_s +blk.35.attn_k_b.weight iq4_nl blk.35.attn_output.weight iq3_s +blk.36.attn_k_b.weight iq4_nl blk.36.attn_output.weight iq3_s +blk.37.attn_k_b.weight iq4_nl blk.37.attn_output.weight iq3_s +blk.38.attn_k_b.weight iq4_nl blk.38.attn_output.weight iq3_s +blk.39.attn_k_b.weight iq4_nl blk.39.attn_output.weight iq3_s +blk.40.attn_k_b.weight iq4_nl blk.40.attn_output.weight iq3_s +blk.41.attn_k_b.weight iq4_nl blk.41.attn_output.weight iq3_s +blk.42.attn_k_b.weight iq4_nl blk.42.attn_output.weight iq3_s +blk.43.attn_k_b.weight iq4_nl blk.43.attn_output.weight iq3_s +blk.44.attn_k_b.weight iq4_nl blk.44.attn_output.weight iq3_s +blk.45.attn_k_b.weight iq4_nl blk.45.attn_output.weight iq3_s +blk.46.attn_k_b.weight iq4_nl blk.46.attn_output.weight iq3_s +blk.47.attn_k_b.weight iq4_nl blk.47.attn_output.weight iq3_s +blk.48.attn_k_b.weight iq4_nl blk.48.attn_output.weight iq3_s +blk.49.attn_k_b.weight iq4_nl blk.49.attn_output.weight iq3_s +blk.50.attn_k_b.weight iq4_nl blk.50.attn_output.weight iq3_s +blk.51.attn_k_b.weight iq4_nl blk.51.attn_output.weight iq3_s +blk.52.attn_k_b.weight iq4_nl blk.52.attn_output.weight iq3_s +blk.53.attn_k_b.weight iq4_nl blk.53.attn_output.weight iq3_s +blk.54.attn_k_b.weight iq4_nl blk.54.attn_output.weight iq3_s +blk.55.attn_k_b.weight iq4_nl blk.55.attn_output.weight iq3_s +blk.56.attn_k_b.weight iq4_nl blk.56.attn_output.weight iq3_s +blk.57.attn_k_b.weight iq4_nl blk.57.attn_output.weight iq3_s +blk.58.attn_k_b.weight iq4_nl blk.58.attn_output.weight iq3_s +blk.59.attn_k_b.weight iq4_nl blk.59.attn_output.weight iq3_s +blk.60.attn_k_b.weight iq4_nl blk.60.attn_output.weight iq3_s [IQ4_XS] iq4_xs output.weight q6_K +blk.0.attn_k_b.weight iq4_nl blk.0.ffn_down.weight q5_K +blk.1.attn_k_b.weight iq4_nl blk.1.ffn_down.weight q5_K +blk.2.attn_k_b.weight iq4_nl blk.2.ffn_down.weight q5_K +blk.3.attn_k_b.weight iq4_nl blk.3.ffn_down_exps.weight q5_K blk.3.ffn_down_shexp.weight q5_K +blk.4.attn_k_b.weight iq4_nl blk.4.ffn_down_exps.weight q5_K blk.4.ffn_down_shexp.weight q5_K +blk.5.attn_k_b.weight iq4_nl blk.5.ffn_down_exps.weight q5_K blk.5.ffn_down_shexp.weight q5_K +blk.6.attn_k_b.weight iq4_nl blk.6.ffn_down_exps.weight q5_K blk.6.ffn_down_shexp.weight q5_K +blk.7.attn_k_b.weight iq4_nl +blk.8.attn_k_b.weight iq4_nl +blk.9.attn_k_b.weight iq4_nl +blk.10.attn_k_b.weight iq4_nl +blk.11.attn_k_b.weight iq4_nl +blk.12.attn_k_b.weight iq4_nl +blk.13.attn_k_b.weight iq4_nl +blk.14.attn_k_b.weight iq4_nl +blk.15.attn_k_b.weight iq4_nl +blk.16.attn_k_b.weight iq4_nl +blk.17.attn_k_b.weight iq4_nl +blk.18.attn_k_b.weight iq4_nl +blk.19.attn_k_b.weight iq4_nl +blk.20.attn_k_b.weight iq4_nl +blk.21.attn_k_b.weight iq4_nl +blk.22.attn_k_b.weight iq4_nl +blk.23.attn_k_b.weight iq4_nl +blk.24.attn_k_b.weight iq4_nl +blk.25.attn_k_b.weight iq4_nl +blk.26.attn_k_b.weight iq4_nl +blk.27.attn_k_b.weight iq4_nl +blk.28.attn_k_b.weight iq4_nl +blk.29.attn_k_b.weight iq4_nl +blk.30.attn_k_b.weight iq4_nl +blk.31.attn_k_b.weight iq4_nl +blk.32.attn_k_b.weight iq4_nl +blk.33.attn_k_b.weight iq4_nl +blk.34.attn_k_b.weight iq4_nl +blk.35.attn_k_b.weight iq4_nl +blk.36.attn_k_b.weight iq4_nl +blk.37.attn_k_b.weight iq4_nl +blk.38.attn_k_b.weight iq4_nl +blk.39.attn_k_b.weight iq4_nl +blk.40.attn_k_b.weight iq4_nl +blk.41.attn_k_b.weight iq4_nl +blk.42.attn_k_b.weight iq4_nl +blk.43.attn_k_b.weight iq4_nl +blk.44.attn_k_b.weight iq4_nl +blk.45.attn_k_b.weight iq4_nl +blk.46.attn_k_b.weight iq4_nl +blk.47.attn_k_b.weight iq4_nl +blk.48.attn_k_b.weight iq4_nl +blk.49.attn_k_b.weight iq4_nl +blk.50.attn_k_b.weight iq4_nl +blk.51.attn_k_b.weight iq4_nl +blk.52.attn_k_b.weight iq4_nl +blk.53.attn_k_b.weight iq4_nl +blk.54.attn_k_b.weight iq4_nl +blk.55.attn_k_b.weight iq4_nl +blk.56.attn_k_b.weight iq4_nl +blk.57.attn_k_b.weight iq4_nl +blk.58.attn_k_b.weight iq4_nl +blk.59.attn_k_b.weight iq4_nl +blk.60.attn_k_b.weight iq4_nl [IQ1_M] iq1_m output.weight q5_K token_embd.weight q2_K +blk.0.attn_k_b.weight iq4_nl blk.0.attn_output.weight iq2_xxs blk.0.ffn_down.weight q2_K +blk.1.attn_k_b.weight iq4_nl blk.1.attn_output.weight iq2_xxs blk.1.ffn_down.weight q2_K +blk.2.attn_k_b.weight iq4_nl blk.2.attn_output.weight iq2_xxs blk.2.ffn_down.weight q2_K +blk.3.attn_k_b.weight iq4_nl blk.3.attn_output.weight iq2_xxs blk.3.ffn_down_exps.weight q2_K blk.3.ffn_down_shexp.weight q2_K +blk.4.attn_k_b.weight iq4_nl blk.4.attn_output.weight iq2_xxs blk.4.ffn_down_exps.weight q2_K blk.4.ffn_down_shexp.weight q2_K +blk.5.attn_k_b.weight iq4_nl blk.5.attn_output.weight iq2_xxs +blk.6.attn_k_b.weight iq4_nl blk.6.attn_output.weight iq2_xxs +blk.7.attn_k_b.weight iq4_nl blk.7.attn_output.weight iq2_xxs +blk.8.attn_k_b.weight iq4_nl blk.8.attn_output.weight iq2_xxs +blk.9.attn_k_b.weight iq4_nl blk.9.attn_output.weight iq2_xxs +blk.10.attn_k_b.weight iq4_nl blk.10.attn_output.weight iq2_xxs +blk.11.attn_k_b.weight iq4_nl blk.11.attn_output.weight iq2_xxs +blk.12.attn_k_b.weight iq4_nl blk.12.attn_output.weight iq2_xxs +blk.13.attn_k_b.weight iq4_nl blk.13.attn_output.weight iq2_xxs +blk.14.attn_k_b.weight iq4_nl blk.14.attn_output.weight iq2_xxs +blk.15.attn_k_b.weight iq4_nl blk.15.attn_output.weight iq2_xxs +blk.16.attn_k_b.weight iq4_nl blk.16.attn_output.weight iq2_xxs +blk.17.attn_k_b.weight iq4_nl blk.17.attn_output.weight iq2_xxs +blk.18.attn_k_b.weight iq4_nl blk.18.attn_output.weight iq2_xxs +blk.19.attn_k_b.weight iq4_nl blk.19.attn_output.weight iq2_xxs +blk.20.attn_k_b.weight iq4_nl blk.20.attn_output.weight iq2_xxs +blk.21.attn_k_b.weight iq4_nl blk.21.attn_output.weight iq2_xxs +blk.22.attn_k_b.weight iq4_nl blk.22.attn_output.weight iq2_xxs +blk.23.attn_k_b.weight iq4_nl blk.23.attn_output.weight iq2_xxs +blk.24.attn_k_b.weight iq4_nl blk.24.attn_output.weight iq2_xxs +blk.25.attn_k_b.weight iq4_nl blk.25.attn_output.weight iq2_xxs +blk.26.attn_k_b.weight iq4_nl blk.26.attn_output.weight iq2_xxs +blk.27.attn_k_b.weight iq4_nl blk.27.attn_output.weight iq2_xxs +blk.28.attn_k_b.weight iq4_nl blk.28.attn_output.weight iq2_xxs +blk.29.attn_k_b.weight iq4_nl blk.29.attn_output.weight iq2_xxs +blk.30.attn_k_b.weight iq4_nl blk.30.attn_output.weight iq2_xxs +blk.31.attn_k_b.weight iq4_nl blk.31.attn_output.weight iq2_xxs +blk.32.attn_k_b.weight iq4_nl blk.32.attn_output.weight iq2_xxs +blk.33.attn_k_b.weight iq4_nl blk.33.attn_output.weight iq2_xxs +blk.34.attn_k_b.weight iq4_nl blk.34.attn_output.weight iq2_xxs +blk.35.attn_k_b.weight iq4_nl blk.35.attn_output.weight iq2_xxs +blk.36.attn_k_b.weight iq4_nl blk.36.attn_output.weight iq2_xxs +blk.37.attn_k_b.weight iq4_nl blk.37.attn_output.weight iq2_xxs +blk.38.attn_k_b.weight iq4_nl blk.38.attn_output.weight iq2_xxs +blk.39.attn_k_b.weight iq4_nl blk.39.attn_output.weight iq2_xxs +blk.40.attn_k_b.weight iq4_nl blk.40.attn_output.weight iq2_xxs +blk.41.attn_k_b.weight iq4_nl blk.41.attn_output.weight iq2_xxs +blk.42.attn_k_b.weight iq4_nl blk.42.attn_output.weight iq2_xxs +blk.43.attn_k_b.weight iq4_nl blk.43.attn_output.weight iq2_xxs +blk.44.attn_k_b.weight iq4_nl blk.44.attn_output.weight iq2_xxs +blk.45.attn_k_b.weight iq4_nl blk.45.attn_output.weight iq2_xxs +blk.46.attn_k_b.weight iq4_nl blk.46.attn_output.weight iq2_xxs +blk.47.attn_k_b.weight iq4_nl blk.47.attn_output.weight iq2_xxs +blk.48.attn_k_b.weight iq4_nl blk.48.attn_output.weight iq2_xxs +blk.49.attn_k_b.weight iq4_nl blk.49.attn_output.weight iq2_xxs +blk.50.attn_k_b.weight iq4_nl blk.50.attn_output.weight iq2_xxs +blk.51.attn_k_b.weight iq4_nl blk.51.attn_output.weight iq2_xxs +blk.52.attn_k_b.weight iq4_nl blk.52.attn_output.weight iq2_xxs +blk.53.attn_k_b.weight iq4_nl blk.53.attn_output.weight iq2_xxs +blk.54.attn_k_b.weight iq4_nl blk.54.attn_output.weight iq2_xxs +blk.55.attn_k_b.weight iq4_nl blk.55.attn_output.weight iq2_xxs +blk.56.attn_k_b.weight iq4_nl blk.56.attn_output.weight iq2_xxs +blk.57.attn_k_b.weight iq4_nl blk.57.attn_output.weight iq2_xxs +blk.58.attn_k_b.weight iq4_nl blk.58.attn_output.weight iq2_xxs +blk.59.attn_k_b.weight iq4_nl blk.59.attn_output.weight iq2_xxs +blk.60.attn_k_b.weight iq4_nl blk.60.attn_output.weight iq2_xxs [BF16] bf16 -output.weight q6_K [TQ1_0] tq1_0 output.weight q6_K token_embd.weight q4_K +blk.0.attn_k_b.weight q4_0 +blk.1.attn_k_b.weight q4_0 +blk.2.attn_k_b.weight q4_0 +blk.3.attn_k_b.weight q4_0 +blk.4.attn_k_b.weight q4_0 +blk.5.attn_k_b.weight q4_0 +blk.6.attn_k_b.weight q4_0 +blk.7.attn_k_b.weight q4_0 +blk.8.attn_k_b.weight q4_0 +blk.9.attn_k_b.weight q4_0 +blk.10.attn_k_b.weight q4_0 +blk.11.attn_k_b.weight q4_0 +blk.12.attn_k_b.weight q4_0 +blk.13.attn_k_b.weight q4_0 +blk.14.attn_k_b.weight q4_0 +blk.15.attn_k_b.weight q4_0 +blk.16.attn_k_b.weight q4_0 +blk.17.attn_k_b.weight q4_0 +blk.18.attn_k_b.weight q4_0 +blk.19.attn_k_b.weight q4_0 +blk.20.attn_k_b.weight q4_0 +blk.21.attn_k_b.weight q4_0 +blk.22.attn_k_b.weight q4_0 +blk.23.attn_k_b.weight q4_0 +blk.24.attn_k_b.weight q4_0 +blk.25.attn_k_b.weight q4_0 +blk.26.attn_k_b.weight q4_0 +blk.27.attn_k_b.weight q4_0 +blk.28.attn_k_b.weight q4_0 +blk.29.attn_k_b.weight q4_0 +blk.30.attn_k_b.weight q4_0 +blk.31.attn_k_b.weight q4_0 +blk.32.attn_k_b.weight q4_0 +blk.33.attn_k_b.weight q4_0 +blk.34.attn_k_b.weight q4_0 +blk.35.attn_k_b.weight q4_0 +blk.36.attn_k_b.weight q4_0 +blk.37.attn_k_b.weight q4_0 +blk.38.attn_k_b.weight q4_0 +blk.39.attn_k_b.weight q4_0 +blk.40.attn_k_b.weight q4_0 +blk.41.attn_k_b.weight q4_0 +blk.42.attn_k_b.weight q4_0 +blk.43.attn_k_b.weight q4_0 +blk.44.attn_k_b.weight q4_0 +blk.45.attn_k_b.weight q4_0 +blk.46.attn_k_b.weight q4_0 +blk.47.attn_k_b.weight q4_0 +blk.48.attn_k_b.weight q4_0 +blk.49.attn_k_b.weight q4_0 +blk.50.attn_k_b.weight q4_0 +blk.51.attn_k_b.weight q4_0 +blk.52.attn_k_b.weight q4_0 +blk.53.attn_k_b.weight q4_0 +blk.54.attn_k_b.weight q4_0 +blk.55.attn_k_b.weight q4_0 +blk.56.attn_k_b.weight q4_0 +blk.57.attn_k_b.weight q4_0 +blk.58.attn_k_b.weight q4_0 +blk.59.attn_k_b.weight q4_0 +blk.60.attn_k_b.weight q4_0 [TQ2_0] tq2_0 output.weight q6_K token_embd.weight q4_K +blk.0.attn_k_b.weight q4_0 +blk.1.attn_k_b.weight q4_0 +blk.2.attn_k_b.weight q4_0 +blk.3.attn_k_b.weight q4_0 +blk.4.attn_k_b.weight q4_0 +blk.5.attn_k_b.weight q4_0 +blk.6.attn_k_b.weight q4_0 +blk.7.attn_k_b.weight q4_0 +blk.8.attn_k_b.weight q4_0 +blk.9.attn_k_b.weight q4_0 +blk.10.attn_k_b.weight q4_0 +blk.11.attn_k_b.weight q4_0 +blk.12.attn_k_b.weight q4_0 +blk.13.attn_k_b.weight q4_0 +blk.14.attn_k_b.weight q4_0 +blk.15.attn_k_b.weight q4_0 +blk.16.attn_k_b.weight q4_0 +blk.17.attn_k_b.weight q4_0 +blk.18.attn_k_b.weight q4_0 +blk.19.attn_k_b.weight q4_0 +blk.20.attn_k_b.weight q4_0 +blk.21.attn_k_b.weight q4_0 +blk.22.attn_k_b.weight q4_0 +blk.23.attn_k_b.weight q4_0 +blk.24.attn_k_b.weight q4_0 +blk.25.attn_k_b.weight q4_0 +blk.26.attn_k_b.weight q4_0 +blk.27.attn_k_b.weight q4_0 +blk.28.attn_k_b.weight q4_0 +blk.29.attn_k_b.weight q4_0 +blk.30.attn_k_b.weight q4_0 +blk.31.attn_k_b.weight q4_0 +blk.32.attn_k_b.weight q4_0 +blk.33.attn_k_b.weight q4_0 +blk.34.attn_k_b.weight q4_0 +blk.35.attn_k_b.weight q4_0 +blk.36.attn_k_b.weight q4_0 +blk.37.attn_k_b.weight q4_0 +blk.38.attn_k_b.weight q4_0 +blk.39.attn_k_b.weight q4_0 +blk.40.attn_k_b.weight q4_0 +blk.41.attn_k_b.weight q4_0 +blk.42.attn_k_b.weight q4_0 +blk.43.attn_k_b.weight q4_0 +blk.44.attn_k_b.weight q4_0 +blk.45.attn_k_b.weight q4_0 +blk.46.attn_k_b.weight q4_0 +blk.47.attn_k_b.weight q4_0 +blk.48.attn_k_b.weight q4_0 +blk.49.attn_k_b.weight q4_0 +blk.50.attn_k_b.weight q4_0 +blk.51.attn_k_b.weight q4_0 +blk.52.attn_k_b.weight q4_0 +blk.53.attn_k_b.weight q4_0 +blk.54.attn_k_b.weight q4_0 +blk.55.attn_k_b.weight q4_0 +blk.56.attn_k_b.weight q4_0 +blk.57.attn_k_b.weight q4_0 +blk.58.attn_k_b.weight q4_0 +blk.59.attn_k_b.weight q4_0 +blk.60.attn_k_b.weight q4_0 [MXFP4_MOE] mxfp4 output.weight q8_0 diff --git a/tests/snapshots/gemma-3-4b-it.schema b/tests/snapshots/gemma-3-4b-it.schema index 1bec284c87..cf5032a24d 100644 --- a/tests/snapshots/gemma-3-4b-it.schema +++ b/tests/snapshots/gemma-3-4b-it.schema @@ -2,10 +2,8 @@ # n_embd=2560, n_ff=10240, n_vocab=262144, n_layer=34, n_head=8, n_head_kv=4 [F32] f32 -token_embd.weight q6_K [F16] f16 -token_embd.weight q6_K [Q4_0] q4_0 token_embd.weight q6_K @@ -1205,7 +1203,6 @@ blk.33.attn_output.weight iq2_xxs blk.33.attn_v.weight q2_K [BF16] bf16 -token_embd.weight q6_K [TQ1_0] tq1_0 token_embd.weight q6_K diff --git a/tests/snapshots/glm-4.6v.schema b/tests/snapshots/glm-4.6v.schema index 841667a1b3..560745c0fd 100644 --- a/tests/snapshots/glm-4.6v.schema +++ b/tests/snapshots/glm-4.6v.schema @@ -2,10 +2,8 @@ # n_embd=4096, n_ff=10944, n_vocab=151552, n_layer=46, n_head=96, n_head_kv=8, n_expert=128 [F32] f32 -output.weight q6_K [F16] f16 -output.weight q6_K [Q4_0] q4_0 output.weight q6_K @@ -22,1446 +20,2242 @@ output.weight q6_K output.weight q6_K [Q2_K] q2_K -blk.0.ffn_down.weight q3_K +blk.0.ffn_down.weight q4_0 blk.0.attn_output.weight q3_K blk.0.attn_v.weight q4_K -blk.1.ffn_down_exps.weight q3_K -blk.1.ffn_down_shexp.weight q3_K +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_down_shexp.weight q4_0 blk.1.attn_output.weight q3_K blk.1.attn_v.weight q4_K -blk.2.ffn_down_exps.weight q3_K -blk.2.ffn_down_shexp.weight q3_K +blk.2.ffn_down_exps.weight q4_0 +blk.2.ffn_down_shexp.weight q4_0 blk.2.attn_output.weight q3_K blk.2.attn_v.weight q4_K -blk.3.ffn_down_exps.weight q3_K -blk.3.ffn_down_shexp.weight q3_K +blk.3.ffn_down_exps.weight q4_0 +blk.3.ffn_down_shexp.weight q4_0 blk.3.attn_output.weight q3_K blk.3.attn_v.weight q4_K -blk.4.ffn_down_exps.weight q3_K -blk.4.ffn_down_shexp.weight q3_K +blk.4.ffn_down_exps.weight q4_0 +blk.4.ffn_down_shexp.weight q4_0 blk.4.attn_output.weight q3_K blk.4.attn_v.weight q4_K -blk.5.ffn_down_exps.weight q3_K -blk.5.ffn_down_shexp.weight q3_K +blk.5.ffn_down_exps.weight q4_0 +blk.5.ffn_down_shexp.weight q4_0 blk.5.attn_output.weight q3_K blk.5.attn_v.weight q4_K -blk.6.ffn_down_exps.weight q3_K -blk.6.ffn_down_shexp.weight q3_K +blk.6.ffn_down_exps.weight q4_0 +blk.6.ffn_down_shexp.weight q4_0 blk.6.attn_output.weight q3_K blk.6.attn_v.weight q4_K -blk.7.ffn_down_exps.weight q3_K -blk.7.ffn_down_shexp.weight q3_K +blk.7.ffn_down_exps.weight q4_0 +blk.7.ffn_down_shexp.weight q4_0 blk.7.attn_output.weight q3_K blk.7.attn_v.weight q4_K -blk.8.ffn_down_exps.weight q3_K -blk.8.ffn_down_shexp.weight q3_K +blk.8.ffn_down_exps.weight q4_0 +blk.8.ffn_down_shexp.weight q4_0 blk.8.attn_output.weight q3_K blk.8.attn_v.weight q4_K -blk.9.ffn_down_exps.weight q3_K -blk.9.ffn_down_shexp.weight q3_K +blk.9.ffn_down_exps.weight q4_0 +blk.9.ffn_down_shexp.weight q4_0 blk.9.attn_output.weight q3_K blk.9.attn_v.weight q4_K -blk.10.ffn_down_exps.weight q3_K -blk.10.ffn_down_shexp.weight q3_K +blk.10.ffn_down_exps.weight q4_0 +blk.10.ffn_down_shexp.weight q4_0 blk.10.attn_output.weight q3_K blk.10.attn_v.weight q4_K -blk.11.ffn_down_exps.weight q3_K -blk.11.ffn_down_shexp.weight q3_K +blk.11.ffn_down_exps.weight q4_0 +blk.11.ffn_down_shexp.weight q4_0 blk.11.attn_output.weight q3_K blk.11.attn_v.weight q4_K -blk.12.ffn_down_exps.weight q3_K -blk.12.ffn_down_shexp.weight q3_K +blk.12.ffn_down_exps.weight q4_0 +blk.12.ffn_down_shexp.weight q4_0 blk.12.attn_output.weight q3_K blk.12.attn_v.weight q4_K -blk.13.ffn_down_exps.weight q3_K -blk.13.ffn_down_shexp.weight q3_K +blk.13.ffn_down_exps.weight q4_0 +blk.13.ffn_down_shexp.weight q4_0 blk.13.attn_output.weight q3_K blk.13.attn_v.weight q4_K -blk.14.ffn_down_exps.weight q3_K -blk.14.ffn_down_shexp.weight q3_K +blk.14.ffn_down_exps.weight q4_0 +blk.14.ffn_down_shexp.weight q4_0 blk.14.attn_output.weight q3_K blk.14.attn_v.weight q4_K -blk.15.ffn_down_exps.weight q3_K -blk.15.ffn_down_shexp.weight q3_K +blk.15.ffn_down_exps.weight q4_0 +blk.15.ffn_down_shexp.weight q4_0 blk.15.attn_output.weight q3_K blk.15.attn_v.weight q4_K -blk.16.ffn_down_exps.weight q3_K -blk.16.ffn_down_shexp.weight q3_K +blk.16.ffn_down_exps.weight q4_0 +blk.16.ffn_down_shexp.weight q4_0 blk.16.attn_output.weight q3_K blk.16.attn_v.weight q4_K -blk.17.ffn_down_exps.weight q3_K -blk.17.ffn_down_shexp.weight q3_K +blk.17.ffn_down_exps.weight q4_0 +blk.17.ffn_down_shexp.weight q4_0 blk.17.attn_output.weight q3_K blk.17.attn_v.weight q4_K -blk.18.ffn_down_exps.weight q3_K -blk.18.ffn_down_shexp.weight q3_K +blk.18.ffn_down_exps.weight q4_0 +blk.18.ffn_down_shexp.weight q4_0 blk.18.attn_output.weight q3_K blk.18.attn_v.weight q4_K -blk.19.ffn_down_exps.weight q3_K -blk.19.ffn_down_shexp.weight q3_K +blk.19.ffn_down_exps.weight q4_0 +blk.19.ffn_down_shexp.weight q4_0 blk.19.attn_output.weight q3_K blk.19.attn_v.weight q4_K -blk.20.ffn_down_exps.weight q3_K -blk.20.ffn_down_shexp.weight q3_K +blk.20.ffn_down_exps.weight q4_0 +blk.20.ffn_down_shexp.weight q4_0 blk.20.attn_output.weight q3_K blk.20.attn_v.weight q4_K -blk.21.ffn_down_exps.weight q3_K -blk.21.ffn_down_shexp.weight q3_K +blk.21.ffn_down_exps.weight q4_0 +blk.21.ffn_down_shexp.weight q4_0 blk.21.attn_output.weight q3_K blk.21.attn_v.weight q4_K -blk.22.ffn_down_exps.weight q3_K -blk.22.ffn_down_shexp.weight q3_K +blk.22.ffn_down_exps.weight q4_0 +blk.22.ffn_down_shexp.weight q4_0 blk.22.attn_output.weight q3_K blk.22.attn_v.weight q4_K -blk.23.ffn_down_exps.weight q3_K -blk.23.ffn_down_shexp.weight q3_K +blk.23.ffn_down_exps.weight q4_0 +blk.23.ffn_down_shexp.weight q4_0 blk.23.attn_output.weight q3_K blk.23.attn_v.weight q4_K -blk.24.ffn_down_exps.weight q3_K -blk.24.ffn_down_shexp.weight q3_K +blk.24.ffn_down_exps.weight q4_0 +blk.24.ffn_down_shexp.weight q4_0 blk.24.attn_output.weight q3_K blk.24.attn_v.weight q4_K -blk.25.ffn_down_exps.weight q3_K -blk.25.ffn_down_shexp.weight q3_K +blk.25.ffn_down_exps.weight q4_0 +blk.25.ffn_down_shexp.weight q4_0 blk.25.attn_output.weight q3_K blk.25.attn_v.weight q4_K -blk.26.ffn_down_exps.weight q3_K -blk.26.ffn_down_shexp.weight q3_K +blk.26.ffn_down_exps.weight q4_0 +blk.26.ffn_down_shexp.weight q4_0 blk.26.attn_output.weight q3_K blk.26.attn_v.weight q4_K -blk.27.ffn_down_exps.weight q3_K -blk.27.ffn_down_shexp.weight q3_K +blk.27.ffn_down_exps.weight q4_0 +blk.27.ffn_down_shexp.weight q4_0 blk.27.attn_output.weight q3_K blk.27.attn_v.weight q4_K -blk.28.ffn_down_exps.weight q3_K -blk.28.ffn_down_shexp.weight q3_K +blk.28.ffn_down_exps.weight q4_0 +blk.28.ffn_down_shexp.weight q4_0 blk.28.attn_output.weight q3_K blk.28.attn_v.weight q4_K -blk.29.ffn_down_exps.weight q3_K -blk.29.ffn_down_shexp.weight q3_K +blk.29.ffn_down_exps.weight q4_0 +blk.29.ffn_down_shexp.weight q4_0 blk.29.attn_output.weight q3_K blk.29.attn_v.weight q4_K -blk.30.ffn_down_exps.weight q3_K -blk.30.ffn_down_shexp.weight q3_K +blk.30.ffn_down_exps.weight q4_0 +blk.30.ffn_down_shexp.weight q4_0 blk.30.attn_output.weight q3_K blk.30.attn_v.weight q4_K -blk.31.ffn_down_exps.weight q3_K -blk.31.ffn_down_shexp.weight q3_K +blk.31.ffn_down_exps.weight q4_0 +blk.31.ffn_down_shexp.weight q4_0 blk.31.attn_output.weight q3_K blk.31.attn_v.weight q4_K -blk.32.ffn_down_exps.weight q3_K -blk.32.ffn_down_shexp.weight q3_K +blk.32.ffn_down_exps.weight q4_0 +blk.32.ffn_down_shexp.weight q4_0 blk.32.attn_output.weight q3_K blk.32.attn_v.weight q4_K -blk.33.ffn_down_exps.weight q3_K -blk.33.ffn_down_shexp.weight q3_K +blk.33.ffn_down_exps.weight q4_0 +blk.33.ffn_down_shexp.weight q4_0 blk.33.attn_output.weight q3_K blk.33.attn_v.weight q4_K -blk.34.ffn_down_exps.weight q3_K +blk.34.ffn_down_exps.weight q4_0 blk.34.attn_output.weight q3_K -blk.34.ffn_down_shexp.weight q3_K +blk.34.ffn_down_shexp.weight q4_0 blk.34.attn_v.weight q4_K -blk.35.ffn_down_exps.weight q3_K -blk.35.ffn_down_shexp.weight q3_K +blk.35.ffn_down_exps.weight q4_0 +blk.35.ffn_down_shexp.weight q4_0 blk.35.attn_output.weight q3_K blk.35.attn_v.weight q4_K -blk.36.ffn_down_exps.weight q3_K -blk.36.ffn_down_shexp.weight q3_K +blk.36.ffn_down_exps.weight q4_0 +blk.36.ffn_down_shexp.weight q4_0 blk.36.attn_output.weight q3_K blk.36.attn_v.weight q4_K -blk.37.ffn_down_exps.weight q3_K -blk.37.ffn_down_shexp.weight q3_K +blk.37.ffn_down_exps.weight q4_0 +blk.37.ffn_down_shexp.weight q4_0 blk.37.attn_output.weight q3_K blk.37.attn_v.weight q4_K -blk.38.ffn_down_exps.weight q3_K -blk.38.ffn_down_shexp.weight q3_K +blk.38.ffn_down_exps.weight q4_0 +blk.38.ffn_down_shexp.weight q4_0 blk.38.attn_output.weight q3_K blk.38.attn_v.weight q4_K -blk.39.ffn_down_exps.weight q3_K -blk.39.ffn_down_shexp.weight q3_K +blk.39.ffn_down_exps.weight q4_0 +blk.39.ffn_down_shexp.weight q4_0 blk.39.attn_output.weight q3_K blk.39.attn_v.weight q4_K -blk.40.ffn_down_exps.weight q3_K -blk.40.ffn_down_shexp.weight q3_K +blk.40.ffn_down_exps.weight q4_0 +blk.40.ffn_down_shexp.weight q4_0 blk.40.attn_output.weight q3_K blk.40.attn_v.weight q4_K -blk.41.ffn_down_exps.weight q3_K -blk.41.ffn_down_shexp.weight q3_K +blk.41.ffn_down_exps.weight q4_0 +blk.41.ffn_down_shexp.weight q4_0 blk.41.attn_output.weight q3_K blk.41.attn_v.weight q4_K -blk.42.ffn_down_exps.weight q3_K -blk.42.ffn_down_shexp.weight q3_K +blk.42.ffn_down_exps.weight q4_0 +blk.42.ffn_down_shexp.weight q4_0 blk.42.attn_v.weight q4_K blk.42.attn_output.weight q3_K -blk.43.ffn_down_exps.weight q3_K -blk.43.ffn_down_shexp.weight q3_K +blk.43.ffn_down_exps.weight q4_0 +blk.43.ffn_down_shexp.weight q4_0 blk.43.attn_output.weight q3_K blk.43.attn_v.weight q4_K -blk.44.ffn_down_exps.weight q3_K -blk.44.ffn_down_shexp.weight q3_K +blk.44.ffn_down_exps.weight q4_0 +blk.44.ffn_down_shexp.weight q4_0 blk.44.attn_output.weight q3_K blk.44.attn_v.weight q4_K output.weight q6_K -blk.45.ffn_down_exps.weight q3_K -blk.45.ffn_down_shexp.weight q3_K +blk.45.ffn_down_exps.weight q4_0 +blk.45.ffn_down_shexp.weight q4_0 blk.45.attn_output.weight q3_K blk.45.attn_v.weight q4_K [Q3_K_S] q3_K +blk.0.ffn_down.weight q4_0 +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_down_shexp.weight q4_0 +blk.2.ffn_down_exps.weight q4_0 +blk.2.ffn_down_shexp.weight q4_0 +blk.3.ffn_down_exps.weight q4_0 +blk.3.ffn_down_shexp.weight q4_0 +blk.4.ffn_down_exps.weight q4_0 +blk.4.ffn_down_shexp.weight q4_0 +blk.5.ffn_down_exps.weight q4_0 +blk.5.ffn_down_shexp.weight q4_0 +blk.6.ffn_down_exps.weight q4_0 +blk.6.ffn_down_shexp.weight q4_0 +blk.7.ffn_down_exps.weight q4_0 +blk.7.ffn_down_shexp.weight q4_0 +blk.8.ffn_down_exps.weight q4_0 +blk.8.ffn_down_shexp.weight q4_0 +blk.9.ffn_down_exps.weight q4_0 +blk.9.ffn_down_shexp.weight q4_0 +blk.10.ffn_down_exps.weight q4_0 +blk.10.ffn_down_shexp.weight q4_0 +blk.11.ffn_down_exps.weight q4_0 +blk.11.ffn_down_shexp.weight q4_0 +blk.12.ffn_down_exps.weight q4_0 +blk.12.ffn_down_shexp.weight q4_0 +blk.13.ffn_down_exps.weight q4_0 +blk.13.ffn_down_shexp.weight q4_0 +blk.14.ffn_down_exps.weight q4_0 +blk.14.ffn_down_shexp.weight q4_0 +blk.15.ffn_down_exps.weight q4_0 +blk.15.ffn_down_shexp.weight q4_0 +blk.16.ffn_down_exps.weight q4_0 +blk.16.ffn_down_shexp.weight q4_0 +blk.17.ffn_down_exps.weight q4_0 +blk.17.ffn_down_shexp.weight q4_0 +blk.18.ffn_down_exps.weight q4_0 +blk.18.ffn_down_shexp.weight q4_0 +blk.19.ffn_down_exps.weight q4_0 +blk.19.ffn_down_shexp.weight q4_0 +blk.20.ffn_down_exps.weight q4_0 +blk.20.ffn_down_shexp.weight q4_0 +blk.21.ffn_down_exps.weight q4_0 +blk.21.ffn_down_shexp.weight q4_0 +blk.22.ffn_down_exps.weight q4_0 +blk.22.ffn_down_shexp.weight q4_0 +blk.23.ffn_down_exps.weight q4_0 +blk.23.ffn_down_shexp.weight q4_0 +blk.24.ffn_down_exps.weight q4_0 +blk.24.ffn_down_shexp.weight q4_0 +blk.25.ffn_down_exps.weight q4_0 +blk.25.ffn_down_shexp.weight q4_0 +blk.26.ffn_down_exps.weight q4_0 +blk.26.ffn_down_shexp.weight q4_0 +blk.27.ffn_down_exps.weight q4_0 +blk.27.ffn_down_shexp.weight q4_0 +blk.28.ffn_down_exps.weight q4_0 +blk.28.ffn_down_shexp.weight q4_0 +blk.29.ffn_down_exps.weight q4_0 +blk.29.ffn_down_shexp.weight q4_0 +blk.30.ffn_down_exps.weight q4_0 +blk.30.ffn_down_shexp.weight q4_0 +blk.31.ffn_down_exps.weight q4_0 +blk.31.ffn_down_shexp.weight q4_0 +blk.32.ffn_down_exps.weight q4_0 +blk.32.ffn_down_shexp.weight q4_0 +blk.33.ffn_down_exps.weight q4_0 +blk.33.ffn_down_shexp.weight q4_0 +blk.34.ffn_down_exps.weight q4_0 +blk.34.ffn_down_shexp.weight q4_0 +blk.35.ffn_down_exps.weight q4_0 +blk.35.ffn_down_shexp.weight q4_0 +blk.36.ffn_down_exps.weight q4_0 +blk.36.ffn_down_shexp.weight q4_0 +blk.37.ffn_down_exps.weight q4_0 +blk.37.ffn_down_shexp.weight q4_0 +blk.38.ffn_down_exps.weight q4_0 +blk.38.ffn_down_shexp.weight q4_0 +blk.39.ffn_down_exps.weight q4_0 +blk.39.ffn_down_shexp.weight q4_0 +blk.40.ffn_down_exps.weight q4_0 +blk.40.ffn_down_shexp.weight q4_0 +blk.41.ffn_down_exps.weight q4_0 +blk.41.ffn_down_shexp.weight q4_0 +blk.42.ffn_down_exps.weight q4_0 +blk.42.ffn_down_shexp.weight q4_0 +blk.43.ffn_down_exps.weight q4_0 +blk.43.ffn_down_shexp.weight q4_0 +blk.44.ffn_down_exps.weight q4_0 +blk.44.ffn_down_shexp.weight q4_0 output.weight q6_K +blk.45.ffn_down_exps.weight q4_0 +blk.45.ffn_down_shexp.weight q4_0 [Q3_K_M] q3_K -blk.0.ffn_down.weight q5_K +blk.0.ffn_down.weight q5_1 blk.0.attn_output.weight q4_K blk.0.attn_v.weight q5_K -blk.1.ffn_down_exps.weight q5_K -blk.1.ffn_down_shexp.weight q5_K +blk.1.ffn_down_exps.weight q5_1 +blk.1.ffn_down_shexp.weight q5_1 blk.1.attn_output.weight q4_K blk.1.attn_v.weight q5_K -blk.2.ffn_down_exps.weight q4_K -blk.2.ffn_down_shexp.weight q4_K +blk.2.ffn_down_exps.weight q5_0 +blk.2.ffn_down_shexp.weight q5_0 blk.2.attn_output.weight q4_K blk.2.attn_v.weight q4_K -blk.3.ffn_down_exps.weight q4_K -blk.3.ffn_down_shexp.weight q4_K +blk.3.ffn_down_exps.weight q5_0 +blk.3.ffn_down_shexp.weight q5_0 blk.3.attn_output.weight q4_K blk.3.attn_v.weight q4_K -blk.4.ffn_down_exps.weight q4_K -blk.4.ffn_down_shexp.weight q4_K +blk.4.ffn_down_exps.weight q5_0 +blk.4.ffn_down_shexp.weight q5_0 blk.4.attn_output.weight q4_K blk.4.attn_v.weight q4_K -blk.5.ffn_down_exps.weight q4_K -blk.5.ffn_down_shexp.weight q4_K +blk.5.ffn_down_exps.weight q5_0 +blk.5.ffn_down_shexp.weight q5_0 blk.5.attn_output.weight q4_K blk.5.attn_v.weight q4_K -blk.6.ffn_down_exps.weight q4_K -blk.6.ffn_down_shexp.weight q4_K +blk.6.ffn_down_exps.weight q5_0 +blk.6.ffn_down_shexp.weight q5_0 blk.6.attn_output.weight q4_K blk.6.attn_v.weight q4_K -blk.7.ffn_down_exps.weight q4_K -blk.7.ffn_down_shexp.weight q4_K +blk.7.ffn_down_exps.weight q5_0 +blk.7.ffn_down_shexp.weight q5_0 blk.7.attn_output.weight q4_K blk.7.attn_v.weight q4_K -blk.8.ffn_down_exps.weight q4_K -blk.8.ffn_down_shexp.weight q4_K +blk.8.ffn_down_exps.weight q5_0 +blk.8.ffn_down_shexp.weight q5_0 blk.8.attn_output.weight q4_K blk.8.attn_v.weight q4_K -blk.9.ffn_down_exps.weight q4_K -blk.9.ffn_down_shexp.weight q4_K +blk.9.ffn_down_exps.weight q5_0 +blk.9.ffn_down_shexp.weight q5_0 blk.9.attn_output.weight q4_K blk.9.attn_v.weight q4_K -blk.10.ffn_down_exps.weight q4_K -blk.10.ffn_down_shexp.weight q4_K +blk.10.ffn_down_exps.weight q5_0 +blk.10.ffn_down_shexp.weight q5_0 blk.10.attn_output.weight q4_K blk.10.attn_v.weight q4_K -blk.11.ffn_down_exps.weight q4_K -blk.11.ffn_down_shexp.weight q4_K +blk.11.ffn_down_exps.weight q5_0 +blk.11.ffn_down_shexp.weight q5_0 blk.11.attn_output.weight q4_K blk.11.attn_v.weight q4_K -blk.12.ffn_down_exps.weight q4_K -blk.12.ffn_down_shexp.weight q4_K +blk.12.ffn_down_exps.weight q5_0 +blk.12.ffn_down_shexp.weight q5_0 blk.12.attn_output.weight q4_K blk.12.attn_v.weight q4_K -blk.13.ffn_down_exps.weight q4_K -blk.13.ffn_down_shexp.weight q4_K +blk.13.ffn_down_exps.weight q5_0 +blk.13.ffn_down_shexp.weight q5_0 blk.13.attn_output.weight q4_K blk.13.attn_v.weight q4_K -blk.14.ffn_down_exps.weight q4_K -blk.14.ffn_down_shexp.weight q4_K +blk.14.ffn_down_exps.weight q5_0 +blk.14.ffn_down_shexp.weight q5_0 blk.14.attn_output.weight q4_K blk.14.attn_v.weight q4_K -blk.15.ffn_down_exps.weight q4_K -blk.15.ffn_down_shexp.weight q4_K +blk.15.ffn_down_exps.weight q5_0 +blk.15.ffn_down_shexp.weight q5_0 blk.15.attn_output.weight q4_K blk.15.attn_v.weight q4_K -blk.16.ffn_down_exps.weight q4_K -blk.16.ffn_down_shexp.weight q4_K +blk.16.ffn_down_exps.weight q5_0 +blk.16.ffn_down_shexp.weight q5_0 blk.16.attn_output.weight q4_K blk.16.attn_v.weight q4_K -blk.17.ffn_down_exps.weight q4_K -blk.17.ffn_down_shexp.weight q4_K +blk.17.ffn_down_exps.weight q5_0 +blk.17.ffn_down_shexp.weight q5_0 blk.17.attn_output.weight q4_K blk.17.attn_v.weight q4_K -blk.18.ffn_down_exps.weight q4_K -blk.18.ffn_down_shexp.weight q4_K +blk.18.ffn_down_exps.weight q5_0 +blk.18.ffn_down_shexp.weight q5_0 blk.18.attn_output.weight q4_K blk.18.attn_v.weight q4_K -blk.19.ffn_down_exps.weight q4_K -blk.19.ffn_down_shexp.weight q4_K +blk.19.ffn_down_exps.weight q5_0 +blk.19.ffn_down_shexp.weight q5_0 blk.19.attn_output.weight q4_K blk.19.attn_v.weight q4_K -blk.20.ffn_down_exps.weight q4_K -blk.20.ffn_down_shexp.weight q4_K +blk.20.ffn_down_exps.weight q5_0 +blk.20.ffn_down_shexp.weight q5_0 blk.20.attn_output.weight q4_K blk.20.attn_v.weight q4_K -blk.21.ffn_down_exps.weight q4_K -blk.21.ffn_down_shexp.weight q4_K +blk.21.ffn_down_exps.weight q5_0 +blk.21.ffn_down_shexp.weight q5_0 blk.21.attn_output.weight q4_K blk.21.attn_v.weight q4_K -blk.22.ffn_down_exps.weight q4_K -blk.22.ffn_down_shexp.weight q4_K +blk.22.ffn_down_exps.weight q5_0 +blk.22.ffn_down_shexp.weight q5_0 blk.22.attn_output.weight q4_K blk.22.attn_v.weight q4_K -blk.23.ffn_down_exps.weight q4_K -blk.23.ffn_down_shexp.weight q4_K +blk.23.ffn_down_exps.weight q5_0 +blk.23.ffn_down_shexp.weight q5_0 blk.23.attn_output.weight q4_K blk.23.attn_v.weight q4_K -blk.24.ffn_down_exps.weight q4_K -blk.24.ffn_down_shexp.weight q4_K +blk.24.ffn_down_exps.weight q5_0 +blk.24.ffn_down_shexp.weight q5_0 blk.24.attn_output.weight q4_K blk.24.attn_v.weight q4_K -blk.25.ffn_down_exps.weight q4_K -blk.25.ffn_down_shexp.weight q4_K +blk.25.ffn_down_exps.weight q5_0 +blk.25.ffn_down_shexp.weight q5_0 blk.25.attn_output.weight q4_K blk.25.attn_v.weight q4_K -blk.26.ffn_down_exps.weight q4_K -blk.26.ffn_down_shexp.weight q4_K +blk.26.ffn_down_exps.weight q5_0 +blk.26.ffn_down_shexp.weight q5_0 blk.26.attn_output.weight q4_K blk.26.attn_v.weight q4_K -blk.27.ffn_down_exps.weight q4_K -blk.27.ffn_down_shexp.weight q4_K +blk.27.ffn_down_exps.weight q5_0 +blk.27.ffn_down_shexp.weight q5_0 blk.27.attn_output.weight q4_K blk.27.attn_v.weight q4_K -blk.28.ffn_down_exps.weight q4_K -blk.28.ffn_down_shexp.weight q4_K +blk.28.ffn_down_exps.weight q5_0 +blk.28.ffn_down_shexp.weight q5_0 blk.28.attn_output.weight q4_K blk.28.attn_v.weight q4_K -blk.29.ffn_down_exps.weight q4_K -blk.29.ffn_down_shexp.weight q4_K +blk.29.ffn_down_exps.weight q5_0 +blk.29.ffn_down_shexp.weight q5_0 blk.29.attn_output.weight q4_K blk.29.attn_v.weight q4_K -blk.30.ffn_down_exps.weight q4_K -blk.30.ffn_down_shexp.weight q4_K +blk.30.ffn_down_exps.weight q5_0 +blk.30.ffn_down_shexp.weight q5_0 blk.30.attn_output.weight q4_K blk.30.attn_v.weight q4_K -blk.31.ffn_down_exps.weight q4_K -blk.31.ffn_down_shexp.weight q4_K +blk.31.ffn_down_exps.weight q5_0 +blk.31.ffn_down_shexp.weight q5_0 blk.31.attn_output.weight q4_K blk.31.attn_v.weight q4_K -blk.32.ffn_down_exps.weight q4_K -blk.32.ffn_down_shexp.weight q4_K +blk.32.ffn_down_exps.weight q5_0 +blk.32.ffn_down_shexp.weight q5_0 blk.32.attn_output.weight q4_K blk.32.attn_v.weight q4_K -blk.33.ffn_down_exps.weight q4_K -blk.33.ffn_down_shexp.weight q4_K +blk.33.ffn_down_exps.weight q5_0 +blk.33.ffn_down_shexp.weight q5_0 blk.33.attn_output.weight q4_K blk.33.attn_v.weight q4_K -blk.34.ffn_down_exps.weight q4_K +blk.34.ffn_down_exps.weight q5_0 blk.34.attn_output.weight q4_K -blk.34.ffn_down_shexp.weight q4_K +blk.34.ffn_down_shexp.weight q5_0 blk.34.attn_v.weight q4_K -blk.35.ffn_down_exps.weight q4_K -blk.35.ffn_down_shexp.weight q4_K +blk.35.ffn_down_exps.weight q5_0 +blk.35.ffn_down_shexp.weight q5_0 blk.35.attn_output.weight q4_K blk.35.attn_v.weight q4_K -blk.36.ffn_down_exps.weight q4_K -blk.36.ffn_down_shexp.weight q4_K +blk.36.ffn_down_exps.weight q5_0 +blk.36.ffn_down_shexp.weight q5_0 blk.36.attn_output.weight q4_K blk.36.attn_v.weight q4_K -blk.37.ffn_down_exps.weight q4_K -blk.37.ffn_down_shexp.weight q4_K +blk.37.ffn_down_exps.weight q5_0 +blk.37.ffn_down_shexp.weight q5_0 blk.37.attn_output.weight q4_K blk.37.attn_v.weight q4_K -blk.38.ffn_down_exps.weight q4_K -blk.38.ffn_down_shexp.weight q4_K +blk.38.ffn_down_exps.weight q5_0 +blk.38.ffn_down_shexp.weight q5_0 blk.38.attn_output.weight q4_K blk.38.attn_v.weight q4_K -blk.39.ffn_down_exps.weight q4_K -blk.39.ffn_down_shexp.weight q4_K +blk.39.ffn_down_exps.weight q5_0 +blk.39.ffn_down_shexp.weight q5_0 blk.39.attn_output.weight q4_K blk.39.attn_v.weight q4_K -blk.40.ffn_down_exps.weight q4_K -blk.40.ffn_down_shexp.weight q4_K +blk.40.ffn_down_exps.weight q5_0 +blk.40.ffn_down_shexp.weight q5_0 blk.40.attn_output.weight q4_K blk.40.attn_v.weight q4_K -blk.41.ffn_down_exps.weight q4_K -blk.41.ffn_down_shexp.weight q4_K +blk.41.ffn_down_exps.weight q5_0 +blk.41.ffn_down_shexp.weight q5_0 blk.41.attn_output.weight q4_K blk.41.attn_v.weight q4_K -blk.42.ffn_down_exps.weight q4_K -blk.42.ffn_down_shexp.weight q4_K +blk.42.ffn_down_exps.weight q5_0 +blk.42.ffn_down_shexp.weight q5_0 blk.42.attn_v.weight q4_K blk.42.attn_output.weight q4_K -blk.43.ffn_down_exps.weight q4_K -blk.43.ffn_down_shexp.weight q4_K +blk.43.ffn_down_exps.weight q5_0 +blk.43.ffn_down_shexp.weight q5_0 blk.43.attn_output.weight q4_K blk.43.attn_v.weight q4_K -blk.44.ffn_down_exps.weight q4_K -blk.44.ffn_down_shexp.weight q4_K +blk.44.ffn_down_exps.weight q5_0 +blk.44.ffn_down_shexp.weight q5_0 blk.44.attn_output.weight q4_K blk.44.attn_v.weight q4_K output.weight q6_K -blk.45.ffn_down_exps.weight q4_K -blk.45.ffn_down_shexp.weight q4_K +blk.45.ffn_down_exps.weight q5_0 +blk.45.ffn_down_shexp.weight q5_0 blk.45.attn_output.weight q4_K blk.45.attn_v.weight q4_K [Q3_K_L] q3_K -blk.0.ffn_down.weight q5_K +blk.0.ffn_down.weight q5_1 blk.0.attn_output.weight q5_K blk.0.attn_v.weight q5_K -blk.1.ffn_down_exps.weight q5_K -blk.1.ffn_down_shexp.weight q5_K +blk.1.ffn_down_exps.weight q5_1 +blk.1.ffn_down_shexp.weight q5_1 blk.1.attn_output.weight q5_K blk.1.attn_v.weight q5_K -blk.2.ffn_down_exps.weight q5_K -blk.2.ffn_down_shexp.weight q5_K +blk.2.ffn_down_exps.weight q5_1 +blk.2.ffn_down_shexp.weight q5_1 blk.2.attn_output.weight q5_K blk.2.attn_v.weight q5_K -blk.3.ffn_down_exps.weight q5_K -blk.3.ffn_down_shexp.weight q5_K +blk.3.ffn_down_exps.weight q5_1 +blk.3.ffn_down_shexp.weight q5_1 blk.3.attn_output.weight q5_K blk.3.attn_v.weight q5_K -blk.4.ffn_down_exps.weight q5_K -blk.4.ffn_down_shexp.weight q5_K +blk.4.ffn_down_exps.weight q5_1 +blk.4.ffn_down_shexp.weight q5_1 blk.4.attn_output.weight q5_K blk.4.attn_v.weight q5_K -blk.5.ffn_down_exps.weight q5_K -blk.5.ffn_down_shexp.weight q5_K +blk.5.ffn_down_exps.weight q5_1 +blk.5.ffn_down_shexp.weight q5_1 blk.5.attn_output.weight q5_K blk.5.attn_v.weight q5_K -blk.6.ffn_down_exps.weight q5_K -blk.6.ffn_down_shexp.weight q5_K +blk.6.ffn_down_exps.weight q5_1 +blk.6.ffn_down_shexp.weight q5_1 blk.6.attn_output.weight q5_K blk.6.attn_v.weight q5_K -blk.7.ffn_down_exps.weight q5_K -blk.7.ffn_down_shexp.weight q5_K +blk.7.ffn_down_exps.weight q5_1 +blk.7.ffn_down_shexp.weight q5_1 blk.7.attn_output.weight q5_K blk.7.attn_v.weight q5_K -blk.8.ffn_down_exps.weight q5_K -blk.8.ffn_down_shexp.weight q5_K +blk.8.ffn_down_exps.weight q5_1 +blk.8.ffn_down_shexp.weight q5_1 blk.8.attn_output.weight q5_K blk.8.attn_v.weight q5_K -blk.9.ffn_down_exps.weight q5_K -blk.9.ffn_down_shexp.weight q5_K +blk.9.ffn_down_exps.weight q5_1 +blk.9.ffn_down_shexp.weight q5_1 blk.9.attn_output.weight q5_K blk.9.attn_v.weight q5_K -blk.10.ffn_down_exps.weight q5_K -blk.10.ffn_down_shexp.weight q5_K +blk.10.ffn_down_exps.weight q5_1 +blk.10.ffn_down_shexp.weight q5_1 blk.10.attn_output.weight q5_K blk.10.attn_v.weight q5_K -blk.11.ffn_down_exps.weight q5_K -blk.11.ffn_down_shexp.weight q5_K +blk.11.ffn_down_exps.weight q5_1 +blk.11.ffn_down_shexp.weight q5_1 blk.11.attn_output.weight q5_K blk.11.attn_v.weight q5_K -blk.12.ffn_down_exps.weight q5_K -blk.12.ffn_down_shexp.weight q5_K +blk.12.ffn_down_exps.weight q5_1 +blk.12.ffn_down_shexp.weight q5_1 blk.12.attn_output.weight q5_K blk.12.attn_v.weight q5_K -blk.13.ffn_down_exps.weight q5_K -blk.13.ffn_down_shexp.weight q5_K +blk.13.ffn_down_exps.weight q5_1 +blk.13.ffn_down_shexp.weight q5_1 blk.13.attn_output.weight q5_K blk.13.attn_v.weight q5_K -blk.14.ffn_down_exps.weight q5_K -blk.14.ffn_down_shexp.weight q5_K +blk.14.ffn_down_exps.weight q5_1 +blk.14.ffn_down_shexp.weight q5_1 blk.14.attn_output.weight q5_K blk.14.attn_v.weight q5_K -blk.15.ffn_down_exps.weight q5_K -blk.15.ffn_down_shexp.weight q5_K +blk.15.ffn_down_exps.weight q5_1 +blk.15.ffn_down_shexp.weight q5_1 blk.15.attn_output.weight q5_K blk.15.attn_v.weight q5_K -blk.16.ffn_down_exps.weight q5_K -blk.16.ffn_down_shexp.weight q5_K +blk.16.ffn_down_exps.weight q5_1 +blk.16.ffn_down_shexp.weight q5_1 blk.16.attn_output.weight q5_K blk.16.attn_v.weight q5_K -blk.17.ffn_down_exps.weight q5_K -blk.17.ffn_down_shexp.weight q5_K +blk.17.ffn_down_exps.weight q5_1 +blk.17.ffn_down_shexp.weight q5_1 blk.17.attn_output.weight q5_K blk.17.attn_v.weight q5_K -blk.18.ffn_down_exps.weight q5_K -blk.18.ffn_down_shexp.weight q5_K +blk.18.ffn_down_exps.weight q5_1 +blk.18.ffn_down_shexp.weight q5_1 blk.18.attn_output.weight q5_K blk.18.attn_v.weight q5_K -blk.19.ffn_down_exps.weight q5_K -blk.19.ffn_down_shexp.weight q5_K +blk.19.ffn_down_exps.weight q5_1 +blk.19.ffn_down_shexp.weight q5_1 blk.19.attn_output.weight q5_K blk.19.attn_v.weight q5_K -blk.20.ffn_down_exps.weight q5_K -blk.20.ffn_down_shexp.weight q5_K +blk.20.ffn_down_exps.weight q5_1 +blk.20.ffn_down_shexp.weight q5_1 blk.20.attn_output.weight q5_K blk.20.attn_v.weight q5_K -blk.21.ffn_down_exps.weight q5_K -blk.21.ffn_down_shexp.weight q5_K +blk.21.ffn_down_exps.weight q5_1 +blk.21.ffn_down_shexp.weight q5_1 blk.21.attn_output.weight q5_K blk.21.attn_v.weight q5_K -blk.22.ffn_down_exps.weight q5_K -blk.22.ffn_down_shexp.weight q5_K +blk.22.ffn_down_exps.weight q5_1 +blk.22.ffn_down_shexp.weight q5_1 blk.22.attn_output.weight q5_K blk.22.attn_v.weight q5_K -blk.23.ffn_down_exps.weight q5_K -blk.23.ffn_down_shexp.weight q5_K +blk.23.ffn_down_exps.weight q5_1 +blk.23.ffn_down_shexp.weight q5_1 blk.23.attn_output.weight q5_K blk.23.attn_v.weight q5_K -blk.24.ffn_down_exps.weight q5_K -blk.24.ffn_down_shexp.weight q5_K +blk.24.ffn_down_exps.weight q5_1 +blk.24.ffn_down_shexp.weight q5_1 blk.24.attn_output.weight q5_K blk.24.attn_v.weight q5_K -blk.25.ffn_down_exps.weight q5_K -blk.25.ffn_down_shexp.weight q5_K +blk.25.ffn_down_exps.weight q5_1 +blk.25.ffn_down_shexp.weight q5_1 blk.25.attn_output.weight q5_K blk.25.attn_v.weight q5_K -blk.26.ffn_down_exps.weight q5_K -blk.26.ffn_down_shexp.weight q5_K +blk.26.ffn_down_exps.weight q5_1 +blk.26.ffn_down_shexp.weight q5_1 blk.26.attn_output.weight q5_K blk.26.attn_v.weight q5_K -blk.27.ffn_down_exps.weight q5_K -blk.27.ffn_down_shexp.weight q5_K +blk.27.ffn_down_exps.weight q5_1 +blk.27.ffn_down_shexp.weight q5_1 blk.27.attn_output.weight q5_K blk.27.attn_v.weight q5_K -blk.28.ffn_down_exps.weight q5_K -blk.28.ffn_down_shexp.weight q5_K +blk.28.ffn_down_exps.weight q5_1 +blk.28.ffn_down_shexp.weight q5_1 blk.28.attn_output.weight q5_K blk.28.attn_v.weight q5_K -blk.29.ffn_down_exps.weight q5_K -blk.29.ffn_down_shexp.weight q5_K +blk.29.ffn_down_exps.weight q5_1 +blk.29.ffn_down_shexp.weight q5_1 blk.29.attn_output.weight q5_K blk.29.attn_v.weight q5_K -blk.30.ffn_down_exps.weight q5_K -blk.30.ffn_down_shexp.weight q5_K +blk.30.ffn_down_exps.weight q5_1 +blk.30.ffn_down_shexp.weight q5_1 blk.30.attn_output.weight q5_K blk.30.attn_v.weight q5_K -blk.31.ffn_down_exps.weight q5_K -blk.31.ffn_down_shexp.weight q5_K +blk.31.ffn_down_exps.weight q5_1 +blk.31.ffn_down_shexp.weight q5_1 blk.31.attn_output.weight q5_K blk.31.attn_v.weight q5_K -blk.32.ffn_down_exps.weight q5_K -blk.32.ffn_down_shexp.weight q5_K +blk.32.ffn_down_exps.weight q5_1 +blk.32.ffn_down_shexp.weight q5_1 blk.32.attn_output.weight q5_K blk.32.attn_v.weight q5_K -blk.33.ffn_down_exps.weight q5_K -blk.33.ffn_down_shexp.weight q5_K +blk.33.ffn_down_exps.weight q5_1 +blk.33.ffn_down_shexp.weight q5_1 blk.33.attn_output.weight q5_K blk.33.attn_v.weight q5_K -blk.34.ffn_down_exps.weight q5_K +blk.34.ffn_down_exps.weight q5_1 blk.34.attn_output.weight q5_K -blk.34.ffn_down_shexp.weight q5_K +blk.34.ffn_down_shexp.weight q5_1 blk.34.attn_v.weight q5_K -blk.35.ffn_down_exps.weight q5_K -blk.35.ffn_down_shexp.weight q5_K +blk.35.ffn_down_exps.weight q5_1 +blk.35.ffn_down_shexp.weight q5_1 blk.35.attn_output.weight q5_K blk.35.attn_v.weight q5_K -blk.36.ffn_down_exps.weight q5_K -blk.36.ffn_down_shexp.weight q5_K +blk.36.ffn_down_exps.weight q5_1 +blk.36.ffn_down_shexp.weight q5_1 blk.36.attn_output.weight q5_K blk.36.attn_v.weight q5_K -blk.37.ffn_down_exps.weight q5_K -blk.37.ffn_down_shexp.weight q5_K +blk.37.ffn_down_exps.weight q5_1 +blk.37.ffn_down_shexp.weight q5_1 blk.37.attn_output.weight q5_K blk.37.attn_v.weight q5_K -blk.38.ffn_down_exps.weight q5_K -blk.38.ffn_down_shexp.weight q5_K +blk.38.ffn_down_exps.weight q5_1 +blk.38.ffn_down_shexp.weight q5_1 blk.38.attn_output.weight q5_K blk.38.attn_v.weight q5_K -blk.39.ffn_down_exps.weight q5_K -blk.39.ffn_down_shexp.weight q5_K +blk.39.ffn_down_exps.weight q5_1 +blk.39.ffn_down_shexp.weight q5_1 blk.39.attn_output.weight q5_K blk.39.attn_v.weight q5_K -blk.40.ffn_down_exps.weight q5_K -blk.40.ffn_down_shexp.weight q5_K +blk.40.ffn_down_exps.weight q5_1 +blk.40.ffn_down_shexp.weight q5_1 blk.40.attn_output.weight q5_K blk.40.attn_v.weight q5_K -blk.41.ffn_down_exps.weight q5_K -blk.41.ffn_down_shexp.weight q5_K +blk.41.ffn_down_exps.weight q5_1 +blk.41.ffn_down_shexp.weight q5_1 blk.41.attn_output.weight q5_K blk.41.attn_v.weight q5_K -blk.42.ffn_down_exps.weight q5_K -blk.42.ffn_down_shexp.weight q5_K +blk.42.ffn_down_exps.weight q5_1 +blk.42.ffn_down_shexp.weight q5_1 blk.42.attn_v.weight q5_K blk.42.attn_output.weight q5_K -blk.43.ffn_down_exps.weight q5_K -blk.43.ffn_down_shexp.weight q5_K +blk.43.ffn_down_exps.weight q5_1 +blk.43.ffn_down_shexp.weight q5_1 blk.43.attn_output.weight q5_K blk.43.attn_v.weight q5_K -blk.44.ffn_down_exps.weight q5_K -blk.44.ffn_down_shexp.weight q5_K +blk.44.ffn_down_exps.weight q5_1 +blk.44.ffn_down_shexp.weight q5_1 blk.44.attn_output.weight q5_K blk.44.attn_v.weight q5_K output.weight q6_K -blk.45.ffn_down_exps.weight q5_K -blk.45.ffn_down_shexp.weight q5_K +blk.45.ffn_down_exps.weight q5_1 +blk.45.ffn_down_shexp.weight q5_1 blk.45.attn_output.weight q5_K blk.45.attn_v.weight q5_K [Q4_K_S] q4_K -blk.0.ffn_down.weight q5_K +blk.0.ffn_down.weight q5_1 blk.0.attn_v.weight q5_K -blk.1.ffn_down_exps.weight q5_K -blk.1.ffn_down_shexp.weight q5_K +blk.1.ffn_down_exps.weight q5_1 +blk.1.ffn_down_shexp.weight q5_1 blk.1.attn_v.weight q5_K -blk.2.ffn_down_exps.weight q5_K -blk.2.ffn_down_shexp.weight q5_K +blk.2.ffn_down_exps.weight q5_1 +blk.2.ffn_down_shexp.weight q5_1 blk.2.attn_v.weight q5_K -blk.3.ffn_down_exps.weight q5_K -blk.3.ffn_down_shexp.weight q5_K +blk.3.ffn_down_exps.weight q5_1 +blk.3.ffn_down_shexp.weight q5_1 blk.3.attn_v.weight q5_K -blk.4.ffn_down_exps.weight q5_K -blk.4.ffn_down_shexp.weight q5_K +blk.4.ffn_down_exps.weight q5_1 +blk.4.ffn_down_shexp.weight q5_1 +blk.5.ffn_down_exps.weight q5_0 +blk.5.ffn_down_shexp.weight q5_0 +blk.6.ffn_down_exps.weight q5_0 +blk.6.ffn_down_shexp.weight q5_0 +blk.7.ffn_down_exps.weight q5_0 +blk.7.ffn_down_shexp.weight q5_0 +blk.8.ffn_down_exps.weight q5_0 +blk.8.ffn_down_shexp.weight q5_0 +blk.9.ffn_down_exps.weight q5_0 +blk.9.ffn_down_shexp.weight q5_0 +blk.10.ffn_down_exps.weight q5_0 +blk.10.ffn_down_shexp.weight q5_0 +blk.11.ffn_down_exps.weight q5_0 +blk.11.ffn_down_shexp.weight q5_0 +blk.12.ffn_down_exps.weight q5_0 +blk.12.ffn_down_shexp.weight q5_0 +blk.13.ffn_down_exps.weight q5_0 +blk.13.ffn_down_shexp.weight q5_0 +blk.14.ffn_down_exps.weight q5_0 +blk.14.ffn_down_shexp.weight q5_0 +blk.15.ffn_down_exps.weight q5_0 +blk.15.ffn_down_shexp.weight q5_0 +blk.16.ffn_down_exps.weight q5_0 +blk.16.ffn_down_shexp.weight q5_0 +blk.17.ffn_down_exps.weight q5_0 +blk.17.ffn_down_shexp.weight q5_0 +blk.18.ffn_down_exps.weight q5_0 +blk.18.ffn_down_shexp.weight q5_0 +blk.19.ffn_down_exps.weight q5_0 +blk.19.ffn_down_shexp.weight q5_0 +blk.20.ffn_down_exps.weight q5_0 +blk.20.ffn_down_shexp.weight q5_0 +blk.21.ffn_down_exps.weight q5_0 +blk.21.ffn_down_shexp.weight q5_0 +blk.22.ffn_down_exps.weight q5_0 +blk.22.ffn_down_shexp.weight q5_0 +blk.23.ffn_down_exps.weight q5_0 +blk.23.ffn_down_shexp.weight q5_0 +blk.24.ffn_down_exps.weight q5_0 +blk.24.ffn_down_shexp.weight q5_0 +blk.25.ffn_down_exps.weight q5_0 +blk.25.ffn_down_shexp.weight q5_0 +blk.26.ffn_down_exps.weight q5_0 +blk.26.ffn_down_shexp.weight q5_0 +blk.27.ffn_down_exps.weight q5_0 +blk.27.ffn_down_shexp.weight q5_0 +blk.28.ffn_down_exps.weight q5_0 +blk.28.ffn_down_shexp.weight q5_0 +blk.29.ffn_down_exps.weight q5_0 +blk.29.ffn_down_shexp.weight q5_0 +blk.30.ffn_down_exps.weight q5_0 +blk.30.ffn_down_shexp.weight q5_0 +blk.31.ffn_down_exps.weight q5_0 +blk.31.ffn_down_shexp.weight q5_0 +blk.32.ffn_down_exps.weight q5_0 +blk.32.ffn_down_shexp.weight q5_0 +blk.33.ffn_down_exps.weight q5_0 +blk.33.ffn_down_shexp.weight q5_0 +blk.34.ffn_down_exps.weight q5_0 +blk.34.ffn_down_shexp.weight q5_0 +blk.35.ffn_down_exps.weight q5_0 +blk.35.ffn_down_shexp.weight q5_0 +blk.36.ffn_down_exps.weight q5_0 +blk.36.ffn_down_shexp.weight q5_0 +blk.37.ffn_down_exps.weight q5_0 +blk.37.ffn_down_shexp.weight q5_0 +blk.38.ffn_down_exps.weight q5_0 +blk.38.ffn_down_shexp.weight q5_0 +blk.39.ffn_down_exps.weight q5_0 +blk.39.ffn_down_shexp.weight q5_0 +blk.40.ffn_down_exps.weight q5_0 +blk.40.ffn_down_shexp.weight q5_0 +blk.41.ffn_down_exps.weight q5_0 +blk.41.ffn_down_shexp.weight q5_0 +blk.42.ffn_down_exps.weight q5_0 +blk.42.ffn_down_shexp.weight q5_0 +blk.43.ffn_down_exps.weight q5_0 +blk.43.ffn_down_shexp.weight q5_0 +blk.44.ffn_down_exps.weight q5_0 +blk.44.ffn_down_shexp.weight q5_0 output.weight q6_K +blk.45.ffn_down_exps.weight q5_0 +blk.45.ffn_down_shexp.weight q5_0 [Q4_K_M] q4_K -blk.0.ffn_down.weight q6_K +blk.0.ffn_down.weight q8_0 blk.0.attn_v.weight q6_K -blk.1.ffn_down_exps.weight q6_K -blk.1.ffn_down_shexp.weight q6_K +blk.1.ffn_down_exps.weight q8_0 +blk.1.ffn_down_shexp.weight q8_0 blk.1.attn_v.weight q6_K -blk.2.ffn_down_exps.weight q6_K -blk.2.ffn_down_shexp.weight q6_K +blk.2.ffn_down_exps.weight q8_0 +blk.2.ffn_down_shexp.weight q8_0 blk.2.attn_v.weight q6_K -blk.3.ffn_down_exps.weight q6_K -blk.3.ffn_down_shexp.weight q6_K +blk.3.ffn_down_exps.weight q8_0 +blk.3.ffn_down_shexp.weight q8_0 blk.3.attn_v.weight q6_K -blk.4.ffn_down_exps.weight q6_K -blk.4.ffn_down_shexp.weight q6_K +blk.4.ffn_down_exps.weight q8_0 +blk.4.ffn_down_shexp.weight q8_0 blk.4.attn_v.weight q6_K -blk.7.ffn_down_exps.weight q6_K -blk.7.ffn_down_shexp.weight q6_K +blk.5.ffn_down_exps.weight q5_0 +blk.5.ffn_down_shexp.weight q5_0 +blk.6.ffn_down_exps.weight q5_0 +blk.6.ffn_down_shexp.weight q5_0 +blk.7.ffn_down_exps.weight q8_0 +blk.7.ffn_down_shexp.weight q8_0 blk.7.attn_v.weight q6_K -blk.10.ffn_down_exps.weight q6_K -blk.10.ffn_down_shexp.weight q6_K +blk.8.ffn_down_exps.weight q5_0 +blk.8.ffn_down_shexp.weight q5_0 +blk.9.ffn_down_exps.weight q5_0 +blk.9.ffn_down_shexp.weight q5_0 +blk.10.ffn_down_exps.weight q8_0 +blk.10.ffn_down_shexp.weight q8_0 blk.10.attn_v.weight q6_K -blk.13.ffn_down_exps.weight q6_K -blk.13.ffn_down_shexp.weight q6_K +blk.11.ffn_down_exps.weight q5_0 +blk.11.ffn_down_shexp.weight q5_0 +blk.12.ffn_down_exps.weight q5_0 +blk.12.ffn_down_shexp.weight q5_0 +blk.13.ffn_down_exps.weight q8_0 +blk.13.ffn_down_shexp.weight q8_0 blk.13.attn_v.weight q6_K -blk.16.ffn_down_exps.weight q6_K -blk.16.ffn_down_shexp.weight q6_K +blk.14.ffn_down_exps.weight q5_0 +blk.14.ffn_down_shexp.weight q5_0 +blk.15.ffn_down_exps.weight q5_0 +blk.15.ffn_down_shexp.weight q5_0 +blk.16.ffn_down_exps.weight q8_0 +blk.16.ffn_down_shexp.weight q8_0 blk.16.attn_v.weight q6_K -blk.19.ffn_down_exps.weight q6_K -blk.19.ffn_down_shexp.weight q6_K +blk.17.ffn_down_exps.weight q5_0 +blk.17.ffn_down_shexp.weight q5_0 +blk.18.ffn_down_exps.weight q5_0 +blk.18.ffn_down_shexp.weight q5_0 +blk.19.ffn_down_exps.weight q8_0 +blk.19.ffn_down_shexp.weight q8_0 blk.19.attn_v.weight q6_K -blk.22.ffn_down_exps.weight q6_K -blk.22.ffn_down_shexp.weight q6_K +blk.20.ffn_down_exps.weight q5_0 +blk.20.ffn_down_shexp.weight q5_0 +blk.21.ffn_down_exps.weight q5_0 +blk.21.ffn_down_shexp.weight q5_0 +blk.22.ffn_down_exps.weight q8_0 +blk.22.ffn_down_shexp.weight q8_0 blk.22.attn_v.weight q6_K -blk.25.ffn_down_exps.weight q6_K -blk.25.ffn_down_shexp.weight q6_K +blk.23.ffn_down_exps.weight q5_0 +blk.23.ffn_down_shexp.weight q5_0 +blk.24.ffn_down_exps.weight q5_0 +blk.24.ffn_down_shexp.weight q5_0 +blk.25.ffn_down_exps.weight q8_0 +blk.25.ffn_down_shexp.weight q8_0 blk.25.attn_v.weight q6_K -blk.28.ffn_down_exps.weight q6_K -blk.28.ffn_down_shexp.weight q6_K +blk.26.ffn_down_exps.weight q5_0 +blk.26.ffn_down_shexp.weight q5_0 +blk.27.ffn_down_exps.weight q5_0 +blk.27.ffn_down_shexp.weight q5_0 +blk.28.ffn_down_exps.weight q8_0 +blk.28.ffn_down_shexp.weight q8_0 blk.28.attn_v.weight q6_K -blk.31.ffn_down_exps.weight q6_K -blk.31.ffn_down_shexp.weight q6_K +blk.29.ffn_down_exps.weight q5_0 +blk.29.ffn_down_shexp.weight q5_0 +blk.30.ffn_down_exps.weight q5_0 +blk.30.ffn_down_shexp.weight q5_0 +blk.31.ffn_down_exps.weight q8_0 +blk.31.ffn_down_shexp.weight q8_0 blk.31.attn_v.weight q6_K -blk.34.ffn_down_exps.weight q6_K -blk.34.ffn_down_shexp.weight q6_K +blk.32.ffn_down_exps.weight q5_0 +blk.32.ffn_down_shexp.weight q5_0 +blk.33.ffn_down_exps.weight q5_0 +blk.33.ffn_down_shexp.weight q5_0 +blk.34.ffn_down_exps.weight q8_0 +blk.34.ffn_down_shexp.weight q8_0 blk.34.attn_v.weight q6_K -blk.37.ffn_down_exps.weight q6_K -blk.37.ffn_down_shexp.weight q6_K +blk.35.ffn_down_exps.weight q5_0 +blk.35.ffn_down_shexp.weight q5_0 +blk.36.ffn_down_exps.weight q5_0 +blk.36.ffn_down_shexp.weight q5_0 +blk.37.ffn_down_exps.weight q8_0 +blk.37.ffn_down_shexp.weight q8_0 blk.37.attn_v.weight q6_K -blk.40.ffn_down_exps.weight q6_K -blk.40.ffn_down_shexp.weight q6_K +blk.38.ffn_down_exps.weight q5_0 +blk.38.ffn_down_shexp.weight q5_0 +blk.39.ffn_down_exps.weight q5_0 +blk.39.ffn_down_shexp.weight q5_0 +blk.40.ffn_down_exps.weight q8_0 +blk.40.ffn_down_shexp.weight q8_0 blk.40.attn_v.weight q6_K -blk.41.ffn_down_exps.weight q6_K -blk.41.ffn_down_shexp.weight q6_K +blk.41.ffn_down_exps.weight q8_0 +blk.41.ffn_down_shexp.weight q8_0 blk.41.attn_v.weight q6_K -blk.42.ffn_down_exps.weight q6_K -blk.42.ffn_down_shexp.weight q6_K +blk.42.ffn_down_exps.weight q8_0 +blk.42.ffn_down_shexp.weight q8_0 blk.42.attn_v.weight q6_K -blk.43.ffn_down_exps.weight q6_K -blk.43.ffn_down_shexp.weight q6_K +blk.43.ffn_down_exps.weight q8_0 +blk.43.ffn_down_shexp.weight q8_0 blk.43.attn_v.weight q6_K -blk.44.ffn_down_exps.weight q6_K -blk.44.ffn_down_shexp.weight q6_K +blk.44.ffn_down_exps.weight q8_0 +blk.44.ffn_down_shexp.weight q8_0 blk.44.attn_v.weight q6_K output.weight q6_K -blk.45.ffn_down_exps.weight q6_K -blk.45.ffn_down_shexp.weight q6_K +blk.45.ffn_down_exps.weight q8_0 +blk.45.ffn_down_shexp.weight q8_0 blk.45.attn_v.weight q6_K [Q5_K_S] q5_K +blk.0.ffn_down.weight q5_1 +blk.1.ffn_down_exps.weight q5_1 +blk.1.ffn_down_shexp.weight q5_1 +blk.2.ffn_down_exps.weight q5_1 +blk.2.ffn_down_shexp.weight q5_1 +blk.3.ffn_down_exps.weight q5_1 +blk.3.ffn_down_shexp.weight q5_1 +blk.4.ffn_down_exps.weight q5_1 +blk.4.ffn_down_shexp.weight q5_1 +blk.5.ffn_down_exps.weight q5_1 +blk.5.ffn_down_shexp.weight q5_1 +blk.6.ffn_down_exps.weight q5_1 +blk.6.ffn_down_shexp.weight q5_1 +blk.7.ffn_down_exps.weight q5_1 +blk.7.ffn_down_shexp.weight q5_1 +blk.8.ffn_down_exps.weight q5_1 +blk.8.ffn_down_shexp.weight q5_1 +blk.9.ffn_down_exps.weight q5_1 +blk.9.ffn_down_shexp.weight q5_1 +blk.10.ffn_down_exps.weight q5_1 +blk.10.ffn_down_shexp.weight q5_1 +blk.11.ffn_down_exps.weight q5_1 +blk.11.ffn_down_shexp.weight q5_1 +blk.12.ffn_down_exps.weight q5_1 +blk.12.ffn_down_shexp.weight q5_1 +blk.13.ffn_down_exps.weight q5_1 +blk.13.ffn_down_shexp.weight q5_1 +blk.14.ffn_down_exps.weight q5_1 +blk.14.ffn_down_shexp.weight q5_1 +blk.15.ffn_down_exps.weight q5_1 +blk.15.ffn_down_shexp.weight q5_1 +blk.16.ffn_down_exps.weight q5_1 +blk.16.ffn_down_shexp.weight q5_1 +blk.17.ffn_down_exps.weight q5_1 +blk.17.ffn_down_shexp.weight q5_1 +blk.18.ffn_down_exps.weight q5_1 +blk.18.ffn_down_shexp.weight q5_1 +blk.19.ffn_down_exps.weight q5_1 +blk.19.ffn_down_shexp.weight q5_1 +blk.20.ffn_down_exps.weight q5_1 +blk.20.ffn_down_shexp.weight q5_1 +blk.21.ffn_down_exps.weight q5_1 +blk.21.ffn_down_shexp.weight q5_1 +blk.22.ffn_down_exps.weight q5_1 +blk.22.ffn_down_shexp.weight q5_1 +blk.23.ffn_down_exps.weight q5_1 +blk.23.ffn_down_shexp.weight q5_1 +blk.24.ffn_down_exps.weight q5_1 +blk.24.ffn_down_shexp.weight q5_1 +blk.25.ffn_down_exps.weight q5_1 +blk.25.ffn_down_shexp.weight q5_1 +blk.26.ffn_down_exps.weight q5_1 +blk.26.ffn_down_shexp.weight q5_1 +blk.27.ffn_down_exps.weight q5_1 +blk.27.ffn_down_shexp.weight q5_1 +blk.28.ffn_down_exps.weight q5_1 +blk.28.ffn_down_shexp.weight q5_1 +blk.29.ffn_down_exps.weight q5_1 +blk.29.ffn_down_shexp.weight q5_1 +blk.30.ffn_down_exps.weight q5_1 +blk.30.ffn_down_shexp.weight q5_1 +blk.31.ffn_down_exps.weight q5_1 +blk.31.ffn_down_shexp.weight q5_1 +blk.32.ffn_down_exps.weight q5_1 +blk.32.ffn_down_shexp.weight q5_1 +blk.33.ffn_down_exps.weight q5_1 +blk.33.ffn_down_shexp.weight q5_1 +blk.34.ffn_down_exps.weight q5_1 +blk.34.ffn_down_shexp.weight q5_1 +blk.35.ffn_down_exps.weight q5_1 +blk.35.ffn_down_shexp.weight q5_1 +blk.36.ffn_down_exps.weight q5_1 +blk.36.ffn_down_shexp.weight q5_1 +blk.37.ffn_down_exps.weight q5_1 +blk.37.ffn_down_shexp.weight q5_1 +blk.38.ffn_down_exps.weight q5_1 +blk.38.ffn_down_shexp.weight q5_1 +blk.39.ffn_down_exps.weight q5_1 +blk.39.ffn_down_shexp.weight q5_1 +blk.40.ffn_down_exps.weight q5_1 +blk.40.ffn_down_shexp.weight q5_1 +blk.41.ffn_down_exps.weight q5_1 +blk.41.ffn_down_shexp.weight q5_1 +blk.42.ffn_down_exps.weight q5_1 +blk.42.ffn_down_shexp.weight q5_1 +blk.43.ffn_down_exps.weight q5_1 +blk.43.ffn_down_shexp.weight q5_1 +blk.44.ffn_down_exps.weight q5_1 +blk.44.ffn_down_shexp.weight q5_1 output.weight q6_K +blk.45.ffn_down_exps.weight q5_1 +blk.45.ffn_down_shexp.weight q5_1 [Q5_K_M] q5_K -blk.0.ffn_down.weight q6_K +blk.0.ffn_down.weight q8_0 blk.0.attn_v.weight q6_K -blk.1.ffn_down_exps.weight q6_K -blk.1.ffn_down_shexp.weight q6_K +blk.1.ffn_down_exps.weight q8_0 +blk.1.ffn_down_shexp.weight q8_0 blk.1.attn_v.weight q6_K -blk.2.ffn_down_exps.weight q6_K -blk.2.ffn_down_shexp.weight q6_K +blk.2.ffn_down_exps.weight q8_0 +blk.2.ffn_down_shexp.weight q8_0 blk.2.attn_v.weight q6_K -blk.3.ffn_down_exps.weight q6_K -blk.3.ffn_down_shexp.weight q6_K +blk.3.ffn_down_exps.weight q8_0 +blk.3.ffn_down_shexp.weight q8_0 blk.3.attn_v.weight q6_K -blk.4.ffn_down_exps.weight q6_K -blk.4.ffn_down_shexp.weight q6_K +blk.4.ffn_down_exps.weight q8_0 +blk.4.ffn_down_shexp.weight q8_0 blk.4.attn_v.weight q6_K -blk.7.ffn_down_exps.weight q6_K -blk.7.ffn_down_shexp.weight q6_K +blk.5.ffn_down_exps.weight q5_1 +blk.5.ffn_down_shexp.weight q5_1 +blk.6.ffn_down_exps.weight q5_1 +blk.6.ffn_down_shexp.weight q5_1 +blk.7.ffn_down_exps.weight q8_0 +blk.7.ffn_down_shexp.weight q8_0 blk.7.attn_v.weight q6_K -blk.10.ffn_down_exps.weight q6_K -blk.10.ffn_down_shexp.weight q6_K +blk.8.ffn_down_exps.weight q5_1 +blk.8.ffn_down_shexp.weight q5_1 +blk.9.ffn_down_exps.weight q5_1 +blk.9.ffn_down_shexp.weight q5_1 +blk.10.ffn_down_exps.weight q8_0 +blk.10.ffn_down_shexp.weight q8_0 blk.10.attn_v.weight q6_K -blk.13.ffn_down_exps.weight q6_K -blk.13.ffn_down_shexp.weight q6_K +blk.11.ffn_down_exps.weight q5_1 +blk.11.ffn_down_shexp.weight q5_1 +blk.12.ffn_down_exps.weight q5_1 +blk.12.ffn_down_shexp.weight q5_1 +blk.13.ffn_down_exps.weight q8_0 +blk.13.ffn_down_shexp.weight q8_0 blk.13.attn_v.weight q6_K -blk.16.ffn_down_exps.weight q6_K -blk.16.ffn_down_shexp.weight q6_K +blk.14.ffn_down_exps.weight q5_1 +blk.14.ffn_down_shexp.weight q5_1 +blk.15.ffn_down_exps.weight q5_1 +blk.15.ffn_down_shexp.weight q5_1 +blk.16.ffn_down_exps.weight q8_0 +blk.16.ffn_down_shexp.weight q8_0 blk.16.attn_v.weight q6_K -blk.19.ffn_down_exps.weight q6_K -blk.19.ffn_down_shexp.weight q6_K +blk.17.ffn_down_exps.weight q5_1 +blk.17.ffn_down_shexp.weight q5_1 +blk.18.ffn_down_exps.weight q5_1 +blk.18.ffn_down_shexp.weight q5_1 +blk.19.ffn_down_exps.weight q8_0 +blk.19.ffn_down_shexp.weight q8_0 blk.19.attn_v.weight q6_K -blk.22.ffn_down_exps.weight q6_K -blk.22.ffn_down_shexp.weight q6_K +blk.20.ffn_down_exps.weight q5_1 +blk.20.ffn_down_shexp.weight q5_1 +blk.21.ffn_down_exps.weight q5_1 +blk.21.ffn_down_shexp.weight q5_1 +blk.22.ffn_down_exps.weight q8_0 +blk.22.ffn_down_shexp.weight q8_0 blk.22.attn_v.weight q6_K -blk.25.ffn_down_exps.weight q6_K -blk.25.ffn_down_shexp.weight q6_K +blk.23.ffn_down_exps.weight q5_1 +blk.23.ffn_down_shexp.weight q5_1 +blk.24.ffn_down_exps.weight q5_1 +blk.24.ffn_down_shexp.weight q5_1 +blk.25.ffn_down_exps.weight q8_0 +blk.25.ffn_down_shexp.weight q8_0 blk.25.attn_v.weight q6_K -blk.28.ffn_down_exps.weight q6_K -blk.28.ffn_down_shexp.weight q6_K +blk.26.ffn_down_exps.weight q5_1 +blk.26.ffn_down_shexp.weight q5_1 +blk.27.ffn_down_exps.weight q5_1 +blk.27.ffn_down_shexp.weight q5_1 +blk.28.ffn_down_exps.weight q8_0 +blk.28.ffn_down_shexp.weight q8_0 blk.28.attn_v.weight q6_K -blk.31.ffn_down_exps.weight q6_K -blk.31.ffn_down_shexp.weight q6_K +blk.29.ffn_down_exps.weight q5_1 +blk.29.ffn_down_shexp.weight q5_1 +blk.30.ffn_down_exps.weight q5_1 +blk.30.ffn_down_shexp.weight q5_1 +blk.31.ffn_down_exps.weight q8_0 +blk.31.ffn_down_shexp.weight q8_0 blk.31.attn_v.weight q6_K -blk.34.ffn_down_exps.weight q6_K -blk.34.ffn_down_shexp.weight q6_K +blk.32.ffn_down_exps.weight q5_1 +blk.32.ffn_down_shexp.weight q5_1 +blk.33.ffn_down_exps.weight q5_1 +blk.33.ffn_down_shexp.weight q5_1 +blk.34.ffn_down_exps.weight q8_0 +blk.34.ffn_down_shexp.weight q8_0 blk.34.attn_v.weight q6_K -blk.37.ffn_down_exps.weight q6_K -blk.37.ffn_down_shexp.weight q6_K +blk.35.ffn_down_exps.weight q5_1 +blk.35.ffn_down_shexp.weight q5_1 +blk.36.ffn_down_exps.weight q5_1 +blk.36.ffn_down_shexp.weight q5_1 +blk.37.ffn_down_exps.weight q8_0 +blk.37.ffn_down_shexp.weight q8_0 blk.37.attn_v.weight q6_K -blk.40.ffn_down_exps.weight q6_K -blk.40.ffn_down_shexp.weight q6_K +blk.38.ffn_down_exps.weight q5_1 +blk.38.ffn_down_shexp.weight q5_1 +blk.39.ffn_down_exps.weight q5_1 +blk.39.ffn_down_shexp.weight q5_1 +blk.40.ffn_down_exps.weight q8_0 +blk.40.ffn_down_shexp.weight q8_0 blk.40.attn_v.weight q6_K -blk.41.ffn_down_exps.weight q6_K -blk.41.ffn_down_shexp.weight q6_K +blk.41.ffn_down_exps.weight q8_0 +blk.41.ffn_down_shexp.weight q8_0 blk.41.attn_v.weight q6_K -blk.42.ffn_down_exps.weight q6_K -blk.42.ffn_down_shexp.weight q6_K +blk.42.ffn_down_exps.weight q8_0 +blk.42.ffn_down_shexp.weight q8_0 blk.42.attn_v.weight q6_K -blk.43.ffn_down_exps.weight q6_K -blk.43.ffn_down_shexp.weight q6_K +blk.43.ffn_down_exps.weight q8_0 +blk.43.ffn_down_shexp.weight q8_0 blk.43.attn_v.weight q6_K -blk.44.ffn_down_exps.weight q6_K -blk.44.ffn_down_shexp.weight q6_K +blk.44.ffn_down_exps.weight q8_0 +blk.44.ffn_down_shexp.weight q8_0 blk.44.attn_v.weight q6_K output.weight q6_K -blk.45.ffn_down_exps.weight q6_K -blk.45.ffn_down_shexp.weight q6_K +blk.45.ffn_down_exps.weight q8_0 +blk.45.ffn_down_shexp.weight q8_0 blk.45.attn_v.weight q6_K [Q6_K] q6_K +blk.0.ffn_down.weight q8_0 +blk.1.ffn_down_exps.weight q8_0 +blk.1.ffn_down_shexp.weight q8_0 +blk.2.ffn_down_exps.weight q8_0 +blk.2.ffn_down_shexp.weight q8_0 +blk.3.ffn_down_exps.weight q8_0 +blk.3.ffn_down_shexp.weight q8_0 +blk.4.ffn_down_exps.weight q8_0 +blk.4.ffn_down_shexp.weight q8_0 +blk.5.ffn_down_exps.weight q8_0 +blk.5.ffn_down_shexp.weight q8_0 +blk.6.ffn_down_exps.weight q8_0 +blk.6.ffn_down_shexp.weight q8_0 +blk.7.ffn_down_exps.weight q8_0 +blk.7.ffn_down_shexp.weight q8_0 +blk.8.ffn_down_exps.weight q8_0 +blk.8.ffn_down_shexp.weight q8_0 +blk.9.ffn_down_exps.weight q8_0 +blk.9.ffn_down_shexp.weight q8_0 +blk.10.ffn_down_exps.weight q8_0 +blk.10.ffn_down_shexp.weight q8_0 +blk.11.ffn_down_exps.weight q8_0 +blk.11.ffn_down_shexp.weight q8_0 +blk.12.ffn_down_exps.weight q8_0 +blk.12.ffn_down_shexp.weight q8_0 +blk.13.ffn_down_exps.weight q8_0 +blk.13.ffn_down_shexp.weight q8_0 +blk.14.ffn_down_exps.weight q8_0 +blk.14.ffn_down_shexp.weight q8_0 +blk.15.ffn_down_exps.weight q8_0 +blk.15.ffn_down_shexp.weight q8_0 +blk.16.ffn_down_exps.weight q8_0 +blk.16.ffn_down_shexp.weight q8_0 +blk.17.ffn_down_exps.weight q8_0 +blk.17.ffn_down_shexp.weight q8_0 +blk.18.ffn_down_exps.weight q8_0 +blk.18.ffn_down_shexp.weight q8_0 +blk.19.ffn_down_exps.weight q8_0 +blk.19.ffn_down_shexp.weight q8_0 +blk.20.ffn_down_exps.weight q8_0 +blk.20.ffn_down_shexp.weight q8_0 +blk.21.ffn_down_exps.weight q8_0 +blk.21.ffn_down_shexp.weight q8_0 +blk.22.ffn_down_exps.weight q8_0 +blk.22.ffn_down_shexp.weight q8_0 +blk.23.ffn_down_exps.weight q8_0 +blk.23.ffn_down_shexp.weight q8_0 +blk.24.ffn_down_exps.weight q8_0 +blk.24.ffn_down_shexp.weight q8_0 +blk.25.ffn_down_exps.weight q8_0 +blk.25.ffn_down_shexp.weight q8_0 +blk.26.ffn_down_exps.weight q8_0 +blk.26.ffn_down_shexp.weight q8_0 +blk.27.ffn_down_exps.weight q8_0 +blk.27.ffn_down_shexp.weight q8_0 +blk.28.ffn_down_exps.weight q8_0 +blk.28.ffn_down_shexp.weight q8_0 +blk.29.ffn_down_exps.weight q8_0 +blk.29.ffn_down_shexp.weight q8_0 +blk.30.ffn_down_exps.weight q8_0 +blk.30.ffn_down_shexp.weight q8_0 +blk.31.ffn_down_exps.weight q8_0 +blk.31.ffn_down_shexp.weight q8_0 +blk.32.ffn_down_exps.weight q8_0 +blk.32.ffn_down_shexp.weight q8_0 +blk.33.ffn_down_exps.weight q8_0 +blk.33.ffn_down_shexp.weight q8_0 +blk.34.ffn_down_exps.weight q8_0 +blk.34.ffn_down_shexp.weight q8_0 +blk.35.ffn_down_exps.weight q8_0 +blk.35.ffn_down_shexp.weight q8_0 +blk.36.ffn_down_exps.weight q8_0 +blk.36.ffn_down_shexp.weight q8_0 +blk.37.ffn_down_exps.weight q8_0 +blk.37.ffn_down_shexp.weight q8_0 +blk.38.ffn_down_exps.weight q8_0 +blk.38.ffn_down_shexp.weight q8_0 +blk.39.ffn_down_exps.weight q8_0 +blk.39.ffn_down_shexp.weight q8_0 +blk.40.ffn_down_exps.weight q8_0 +blk.40.ffn_down_shexp.weight q8_0 +blk.41.ffn_down_exps.weight q8_0 +blk.41.ffn_down_shexp.weight q8_0 +blk.42.ffn_down_exps.weight q8_0 +blk.42.ffn_down_shexp.weight q8_0 +blk.43.ffn_down_exps.weight q8_0 +blk.43.ffn_down_shexp.weight q8_0 +blk.44.ffn_down_exps.weight q8_0 +blk.44.ffn_down_shexp.weight q8_0 +blk.45.ffn_down_exps.weight q8_0 +blk.45.ffn_down_shexp.weight q8_0 [IQ2_XXS] iq2_xxs token_embd.weight q2_K -blk.0.ffn_down.weight q2_K +blk.0.ffn_down.weight q4_0 blk.0.attn_v.weight q4_K -blk.1.ffn_down_exps.weight q2_K -blk.1.ffn_down_shexp.weight q2_K +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_down_shexp.weight q4_0 blk.1.attn_v.weight q4_K -blk.2.ffn_down_exps.weight q2_K -blk.2.ffn_down_shexp.weight q2_K +blk.2.ffn_down_exps.weight q4_0 +blk.2.ffn_down_shexp.weight q4_0 blk.2.attn_v.weight q4_K +blk.3.ffn_down_exps.weight iq4_nl +blk.3.ffn_down_shexp.weight iq4_nl blk.3.attn_v.weight q4_K +blk.4.ffn_down_exps.weight iq4_nl +blk.4.ffn_down_shexp.weight iq4_nl blk.4.attn_v.weight q4_K +blk.5.ffn_down_exps.weight iq4_nl +blk.5.ffn_down_shexp.weight iq4_nl blk.5.attn_v.weight q4_K +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_down_shexp.weight iq4_nl blk.6.attn_v.weight q4_K +blk.7.ffn_down_exps.weight iq4_nl +blk.7.ffn_down_shexp.weight iq4_nl blk.7.attn_v.weight q4_K +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_down_shexp.weight iq4_nl blk.8.attn_v.weight q4_K +blk.9.ffn_down_exps.weight iq4_nl +blk.9.ffn_down_shexp.weight iq4_nl blk.9.attn_v.weight q4_K +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_down_shexp.weight iq4_nl blk.10.attn_v.weight q4_K +blk.11.ffn_down_exps.weight iq4_nl +blk.11.ffn_down_shexp.weight iq4_nl blk.11.attn_v.weight q4_K +blk.12.ffn_down_exps.weight iq4_nl +blk.12.ffn_down_shexp.weight iq4_nl blk.12.attn_v.weight q4_K +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_down_shexp.weight iq4_nl blk.13.attn_v.weight q4_K +blk.14.ffn_down_exps.weight iq4_nl +blk.14.ffn_down_shexp.weight iq4_nl blk.14.attn_v.weight q4_K +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_down_shexp.weight iq4_nl blk.15.attn_v.weight q4_K +blk.16.ffn_down_exps.weight iq4_nl +blk.16.ffn_down_shexp.weight iq4_nl blk.16.attn_v.weight q4_K +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_down_shexp.weight iq4_nl blk.17.attn_v.weight q4_K +blk.18.ffn_down_exps.weight iq4_nl +blk.18.ffn_down_shexp.weight iq4_nl blk.18.attn_v.weight q4_K +blk.19.ffn_down_exps.weight iq4_nl +blk.19.ffn_down_shexp.weight iq4_nl blk.19.attn_v.weight q4_K +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_down_shexp.weight iq4_nl blk.20.attn_v.weight q4_K +blk.21.ffn_down_exps.weight iq4_nl +blk.21.ffn_down_shexp.weight iq4_nl blk.21.attn_v.weight q4_K +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_down_shexp.weight iq4_nl blk.22.attn_v.weight q4_K +blk.23.ffn_down_exps.weight iq4_nl +blk.23.ffn_down_shexp.weight iq4_nl blk.23.attn_v.weight q4_K +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_down_shexp.weight iq4_nl blk.24.attn_v.weight q4_K +blk.25.ffn_down_exps.weight iq4_nl +blk.25.ffn_down_shexp.weight iq4_nl blk.25.attn_v.weight q4_K +blk.26.ffn_down_exps.weight iq4_nl +blk.26.ffn_down_shexp.weight iq4_nl blk.26.attn_v.weight q4_K +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_down_shexp.weight iq4_nl blk.27.attn_v.weight q4_K +blk.28.ffn_down_exps.weight iq4_nl +blk.28.ffn_down_shexp.weight iq4_nl blk.28.attn_v.weight q4_K +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_down_shexp.weight iq4_nl blk.29.attn_v.weight q4_K +blk.30.ffn_down_exps.weight iq4_nl +blk.30.ffn_down_shexp.weight iq4_nl blk.30.attn_v.weight q4_K +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_down_shexp.weight iq4_nl blk.31.attn_v.weight q4_K +blk.32.ffn_down_exps.weight iq4_nl +blk.32.ffn_down_shexp.weight iq4_nl blk.32.attn_v.weight q4_K +blk.33.ffn_down_exps.weight iq4_nl +blk.33.ffn_down_shexp.weight iq4_nl blk.33.attn_v.weight q4_K +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_down_shexp.weight iq4_nl blk.34.attn_v.weight q4_K +blk.35.ffn_down_exps.weight iq4_nl +blk.35.ffn_down_shexp.weight iq4_nl blk.35.attn_v.weight q4_K +blk.36.ffn_down_exps.weight iq4_nl +blk.36.ffn_down_shexp.weight iq4_nl blk.36.attn_v.weight q4_K +blk.37.ffn_down_exps.weight iq4_nl +blk.37.ffn_down_shexp.weight iq4_nl blk.37.attn_v.weight q4_K +blk.38.ffn_down_exps.weight iq4_nl +blk.38.ffn_down_shexp.weight iq4_nl blk.38.attn_v.weight q4_K +blk.39.ffn_down_exps.weight iq4_nl +blk.39.ffn_down_shexp.weight iq4_nl blk.39.attn_v.weight q4_K +blk.40.ffn_down_exps.weight iq4_nl +blk.40.ffn_down_shexp.weight iq4_nl blk.40.attn_v.weight q4_K +blk.41.ffn_down_exps.weight iq4_nl +blk.41.ffn_down_shexp.weight iq4_nl blk.41.attn_v.weight q4_K +blk.42.ffn_down_exps.weight iq4_nl +blk.42.ffn_down_shexp.weight iq4_nl blk.42.attn_v.weight q4_K +blk.43.ffn_down_exps.weight iq4_nl +blk.43.ffn_down_shexp.weight iq4_nl blk.43.attn_v.weight q4_K +blk.44.ffn_down_exps.weight iq4_nl +blk.44.ffn_down_shexp.weight iq4_nl blk.44.attn_v.weight q4_K output.weight q5_K +blk.45.ffn_down_exps.weight iq4_nl +blk.45.ffn_down_shexp.weight iq4_nl blk.45.attn_v.weight q4_K [IQ2_XS] iq2_xs token_embd.weight q2_K -blk.0.ffn_down.weight q2_K +blk.0.ffn_down.weight q4_0 blk.0.attn_v.weight q4_K -blk.1.ffn_down_exps.weight q2_K -blk.1.ffn_down_shexp.weight q2_K +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_down_shexp.weight q4_0 blk.1.attn_v.weight q4_K -blk.2.ffn_down_exps.weight q2_K -blk.2.ffn_down_shexp.weight q2_K +blk.2.ffn_down_exps.weight q4_0 +blk.2.ffn_down_shexp.weight q4_0 blk.2.attn_v.weight q4_K +blk.3.ffn_down_exps.weight iq4_nl +blk.3.ffn_down_shexp.weight iq4_nl blk.3.attn_v.weight q4_K +blk.4.ffn_down_exps.weight iq4_nl +blk.4.ffn_down_shexp.weight iq4_nl blk.4.attn_v.weight q4_K +blk.5.ffn_down_exps.weight iq4_nl +blk.5.ffn_down_shexp.weight iq4_nl blk.5.attn_v.weight q4_K +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_down_shexp.weight iq4_nl blk.6.attn_v.weight q4_K +blk.7.ffn_down_exps.weight iq4_nl +blk.7.ffn_down_shexp.weight iq4_nl blk.7.attn_v.weight q4_K +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_down_shexp.weight iq4_nl blk.8.attn_v.weight q4_K +blk.9.ffn_down_exps.weight iq4_nl +blk.9.ffn_down_shexp.weight iq4_nl blk.9.attn_v.weight q4_K +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_down_shexp.weight iq4_nl blk.10.attn_v.weight q4_K +blk.11.ffn_down_exps.weight iq4_nl +blk.11.ffn_down_shexp.weight iq4_nl blk.11.attn_v.weight q4_K +blk.12.ffn_down_exps.weight iq4_nl +blk.12.ffn_down_shexp.weight iq4_nl blk.12.attn_v.weight q4_K +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_down_shexp.weight iq4_nl blk.13.attn_v.weight q4_K +blk.14.ffn_down_exps.weight iq4_nl +blk.14.ffn_down_shexp.weight iq4_nl blk.14.attn_v.weight q4_K +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_down_shexp.weight iq4_nl blk.15.attn_v.weight q4_K +blk.16.ffn_down_exps.weight iq4_nl +blk.16.ffn_down_shexp.weight iq4_nl blk.16.attn_v.weight q4_K +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_down_shexp.weight iq4_nl blk.17.attn_v.weight q4_K +blk.18.ffn_down_exps.weight iq4_nl +blk.18.ffn_down_shexp.weight iq4_nl blk.18.attn_v.weight q4_K +blk.19.ffn_down_exps.weight iq4_nl +blk.19.ffn_down_shexp.weight iq4_nl blk.19.attn_v.weight q4_K +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_down_shexp.weight iq4_nl blk.20.attn_v.weight q4_K +blk.21.ffn_down_exps.weight iq4_nl +blk.21.ffn_down_shexp.weight iq4_nl blk.21.attn_v.weight q4_K +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_down_shexp.weight iq4_nl blk.22.attn_v.weight q4_K +blk.23.ffn_down_exps.weight iq4_nl +blk.23.ffn_down_shexp.weight iq4_nl blk.23.attn_v.weight q4_K +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_down_shexp.weight iq4_nl blk.24.attn_v.weight q4_K +blk.25.ffn_down_exps.weight iq4_nl +blk.25.ffn_down_shexp.weight iq4_nl blk.25.attn_v.weight q4_K +blk.26.ffn_down_exps.weight iq4_nl +blk.26.ffn_down_shexp.weight iq4_nl blk.26.attn_v.weight q4_K +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_down_shexp.weight iq4_nl blk.27.attn_v.weight q4_K +blk.28.ffn_down_exps.weight iq4_nl +blk.28.ffn_down_shexp.weight iq4_nl blk.28.attn_v.weight q4_K +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_down_shexp.weight iq4_nl blk.29.attn_v.weight q4_K +blk.30.ffn_down_exps.weight iq4_nl +blk.30.ffn_down_shexp.weight iq4_nl blk.30.attn_v.weight q4_K +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_down_shexp.weight iq4_nl blk.31.attn_v.weight q4_K +blk.32.ffn_down_exps.weight iq4_nl +blk.32.ffn_down_shexp.weight iq4_nl blk.32.attn_v.weight q4_K +blk.33.ffn_down_exps.weight iq4_nl +blk.33.ffn_down_shexp.weight iq4_nl blk.33.attn_v.weight q4_K +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_down_shexp.weight iq4_nl blk.34.attn_v.weight q4_K +blk.35.ffn_down_exps.weight iq4_nl +blk.35.ffn_down_shexp.weight iq4_nl blk.35.attn_v.weight q4_K +blk.36.ffn_down_exps.weight iq4_nl +blk.36.ffn_down_shexp.weight iq4_nl blk.36.attn_v.weight q4_K +blk.37.ffn_down_exps.weight iq4_nl +blk.37.ffn_down_shexp.weight iq4_nl blk.37.attn_v.weight q4_K +blk.38.ffn_down_exps.weight iq4_nl +blk.38.ffn_down_shexp.weight iq4_nl blk.38.attn_v.weight q4_K +blk.39.ffn_down_exps.weight iq4_nl +blk.39.ffn_down_shexp.weight iq4_nl blk.39.attn_v.weight q4_K +blk.40.ffn_down_exps.weight iq4_nl +blk.40.ffn_down_shexp.weight iq4_nl blk.40.attn_v.weight q4_K +blk.41.ffn_down_exps.weight iq4_nl +blk.41.ffn_down_shexp.weight iq4_nl blk.41.attn_v.weight q4_K +blk.42.ffn_down_exps.weight iq4_nl +blk.42.ffn_down_shexp.weight iq4_nl blk.42.attn_v.weight q4_K +blk.43.ffn_down_exps.weight iq4_nl +blk.43.ffn_down_shexp.weight iq4_nl blk.43.attn_v.weight q4_K +blk.44.ffn_down_exps.weight iq4_nl +blk.44.ffn_down_shexp.weight iq4_nl blk.44.attn_v.weight q4_K output.weight q5_K +blk.45.ffn_down_exps.weight iq4_nl +blk.45.ffn_down_shexp.weight iq4_nl blk.45.attn_v.weight q4_K [Q2_K_S] q2_K -blk.0.ffn_down.weight q4_K +blk.0.ffn_down.weight q5_0 blk.0.attn_v.weight q4_K -blk.1.ffn_down_exps.weight q4_K -blk.1.ffn_down_shexp.weight q4_K +blk.1.ffn_down_exps.weight q5_0 +blk.1.ffn_down_shexp.weight q5_0 blk.1.attn_v.weight q4_K -blk.2.ffn_down_exps.weight q4_K -blk.2.ffn_down_shexp.weight q4_K +blk.2.ffn_down_exps.weight q5_0 +blk.2.ffn_down_shexp.weight q5_0 blk.2.attn_v.weight q4_K -blk.3.ffn_down_exps.weight q4_K -blk.3.ffn_down_shexp.weight q4_K +blk.3.ffn_down_exps.weight q5_0 +blk.3.ffn_down_shexp.weight q5_0 blk.3.attn_v.weight q4_K -blk.4.ffn_down_exps.weight q4_K -blk.4.ffn_down_shexp.weight q4_K +blk.4.ffn_down_exps.weight q5_0 +blk.4.ffn_down_shexp.weight q5_0 blk.4.attn_v.weight q4_K +blk.5.ffn_down_exps.weight q4_0 +blk.5.ffn_down_shexp.weight q4_0 blk.5.attn_v.weight q4_K +blk.6.ffn_down_exps.weight q4_0 +blk.6.ffn_down_shexp.weight q4_0 blk.6.attn_v.weight q4_K +blk.7.ffn_down_exps.weight q4_0 +blk.7.ffn_down_shexp.weight q4_0 blk.7.attn_v.weight q4_K +blk.8.ffn_down_exps.weight q4_0 +blk.8.ffn_down_shexp.weight q4_0 blk.8.attn_v.weight q4_K +blk.9.ffn_down_exps.weight q4_0 +blk.9.ffn_down_shexp.weight q4_0 blk.9.attn_v.weight q4_K +blk.10.ffn_down_exps.weight q4_0 +blk.10.ffn_down_shexp.weight q4_0 blk.10.attn_v.weight q4_K +blk.11.ffn_down_exps.weight q4_0 +blk.11.ffn_down_shexp.weight q4_0 blk.11.attn_v.weight q4_K +blk.12.ffn_down_exps.weight q4_0 +blk.12.ffn_down_shexp.weight q4_0 blk.12.attn_v.weight q4_K +blk.13.ffn_down_exps.weight q4_0 +blk.13.ffn_down_shexp.weight q4_0 blk.13.attn_v.weight q4_K +blk.14.ffn_down_exps.weight q4_0 +blk.14.ffn_down_shexp.weight q4_0 blk.14.attn_v.weight q4_K +blk.15.ffn_down_exps.weight q4_0 +blk.15.ffn_down_shexp.weight q4_0 blk.15.attn_v.weight q4_K +blk.16.ffn_down_exps.weight q4_0 +blk.16.ffn_down_shexp.weight q4_0 blk.16.attn_v.weight q4_K +blk.17.ffn_down_exps.weight q4_0 +blk.17.ffn_down_shexp.weight q4_0 blk.17.attn_v.weight q4_K +blk.18.ffn_down_exps.weight q4_0 +blk.18.ffn_down_shexp.weight q4_0 blk.18.attn_v.weight q4_K +blk.19.ffn_down_exps.weight q4_0 +blk.19.ffn_down_shexp.weight q4_0 blk.19.attn_v.weight q4_K +blk.20.ffn_down_exps.weight q4_0 +blk.20.ffn_down_shexp.weight q4_0 blk.20.attn_v.weight q4_K +blk.21.ffn_down_exps.weight q4_0 +blk.21.ffn_down_shexp.weight q4_0 blk.21.attn_v.weight q4_K +blk.22.ffn_down_exps.weight q4_0 +blk.22.ffn_down_shexp.weight q4_0 blk.22.attn_v.weight q4_K +blk.23.ffn_down_exps.weight q4_0 +blk.23.ffn_down_shexp.weight q4_0 blk.23.attn_v.weight q4_K +blk.24.ffn_down_exps.weight q4_0 +blk.24.ffn_down_shexp.weight q4_0 blk.24.attn_v.weight q4_K +blk.25.ffn_down_exps.weight q4_0 +blk.25.ffn_down_shexp.weight q4_0 blk.25.attn_v.weight q4_K +blk.26.ffn_down_exps.weight q4_0 +blk.26.ffn_down_shexp.weight q4_0 blk.26.attn_v.weight q4_K +blk.27.ffn_down_exps.weight q4_0 +blk.27.ffn_down_shexp.weight q4_0 blk.27.attn_v.weight q4_K +blk.28.ffn_down_exps.weight q4_0 +blk.28.ffn_down_shexp.weight q4_0 blk.28.attn_v.weight q4_K +blk.29.ffn_down_exps.weight q4_0 +blk.29.ffn_down_shexp.weight q4_0 blk.29.attn_v.weight q4_K +blk.30.ffn_down_exps.weight q4_0 +blk.30.ffn_down_shexp.weight q4_0 blk.30.attn_v.weight q4_K +blk.31.ffn_down_exps.weight q4_0 +blk.31.ffn_down_shexp.weight q4_0 blk.31.attn_v.weight q4_K +blk.32.ffn_down_exps.weight q4_0 +blk.32.ffn_down_shexp.weight q4_0 blk.32.attn_v.weight q4_K +blk.33.ffn_down_exps.weight q4_0 +blk.33.ffn_down_shexp.weight q4_0 blk.33.attn_v.weight q4_K +blk.34.ffn_down_exps.weight q4_0 +blk.34.ffn_down_shexp.weight q4_0 blk.34.attn_v.weight q4_K +blk.35.ffn_down_exps.weight q4_0 +blk.35.ffn_down_shexp.weight q4_0 blk.35.attn_v.weight q4_K +blk.36.ffn_down_exps.weight q4_0 +blk.36.ffn_down_shexp.weight q4_0 blk.36.attn_v.weight q4_K +blk.37.ffn_down_exps.weight q4_0 +blk.37.ffn_down_shexp.weight q4_0 blk.37.attn_v.weight q4_K +blk.38.ffn_down_exps.weight q4_0 +blk.38.ffn_down_shexp.weight q4_0 blk.38.attn_v.weight q4_K +blk.39.ffn_down_exps.weight q4_0 +blk.39.ffn_down_shexp.weight q4_0 blk.39.attn_v.weight q4_K +blk.40.ffn_down_exps.weight q4_0 +blk.40.ffn_down_shexp.weight q4_0 blk.40.attn_v.weight q4_K +blk.41.ffn_down_exps.weight q4_0 +blk.41.ffn_down_shexp.weight q4_0 blk.41.attn_v.weight q4_K +blk.42.ffn_down_exps.weight q4_0 +blk.42.ffn_down_shexp.weight q4_0 blk.42.attn_v.weight q4_K +blk.43.ffn_down_exps.weight q4_0 +blk.43.ffn_down_shexp.weight q4_0 blk.43.attn_v.weight q4_K +blk.44.ffn_down_exps.weight q4_0 +blk.44.ffn_down_shexp.weight q4_0 blk.44.attn_v.weight q4_K output.weight q6_K +blk.45.ffn_down_exps.weight q4_0 +blk.45.ffn_down_shexp.weight q4_0 blk.45.attn_v.weight q4_K [IQ3_XS] iq3_s +blk.0.ffn_down.weight iq4_nl blk.0.attn_k.weight iq3_xxs blk.0.attn_q.weight iq3_xxs blk.0.attn_v.weight q4_K +blk.1.ffn_down_exps.weight iq4_nl +blk.1.ffn_down_shexp.weight iq4_nl blk.1.attn_k.weight iq3_xxs blk.1.attn_q.weight iq3_xxs blk.1.attn_v.weight q4_K +blk.2.ffn_down_exps.weight iq4_nl +blk.2.ffn_down_shexp.weight iq4_nl blk.2.attn_k.weight iq3_xxs blk.2.attn_q.weight iq3_xxs blk.2.attn_v.weight q4_K +blk.3.ffn_down_exps.weight iq4_nl +blk.3.ffn_down_shexp.weight iq4_nl blk.3.attn_k.weight iq3_xxs blk.3.attn_q.weight iq3_xxs blk.3.attn_v.weight q4_K +blk.4.ffn_down_exps.weight iq4_nl +blk.4.ffn_down_shexp.weight iq4_nl blk.4.attn_k.weight iq3_xxs blk.4.attn_q.weight iq3_xxs blk.4.attn_v.weight q4_K +blk.5.ffn_down_exps.weight iq4_nl blk.5.ffn_gate_exps.weight iq3_xxs blk.5.ffn_up_exps.weight iq3_xxs +blk.5.ffn_down_shexp.weight iq4_nl blk.5.ffn_gate_shexp.weight iq3_xxs blk.5.ffn_up_shexp.weight iq3_xxs blk.5.attn_k.weight iq3_xxs blk.5.attn_q.weight iq3_xxs blk.5.attn_v.weight q4_K +blk.6.ffn_down_exps.weight iq4_nl blk.6.ffn_gate_exps.weight iq3_xxs blk.6.ffn_up_exps.weight iq3_xxs +blk.6.ffn_down_shexp.weight iq4_nl blk.6.ffn_gate_shexp.weight iq3_xxs blk.6.ffn_up_shexp.weight iq3_xxs blk.6.attn_k.weight iq3_xxs blk.6.attn_q.weight iq3_xxs blk.6.attn_v.weight q4_K +blk.7.ffn_down_exps.weight iq4_nl blk.7.ffn_gate_exps.weight iq3_xxs blk.7.ffn_up_exps.weight iq3_xxs +blk.7.ffn_down_shexp.weight iq4_nl blk.7.ffn_gate_shexp.weight iq3_xxs blk.7.ffn_up_shexp.weight iq3_xxs blk.7.attn_k.weight iq3_xxs blk.7.attn_q.weight iq3_xxs blk.7.attn_v.weight q4_K +blk.8.ffn_down_exps.weight iq4_nl blk.8.ffn_gate_exps.weight iq3_xxs blk.8.ffn_up_exps.weight iq3_xxs +blk.8.ffn_down_shexp.weight iq4_nl blk.8.ffn_gate_shexp.weight iq3_xxs blk.8.ffn_up_shexp.weight iq3_xxs blk.8.attn_k.weight iq3_xxs blk.8.attn_q.weight iq3_xxs blk.8.attn_v.weight q4_K +blk.9.ffn_down_exps.weight iq4_nl blk.9.ffn_gate_exps.weight iq3_xxs blk.9.ffn_up_exps.weight iq3_xxs +blk.9.ffn_down_shexp.weight iq4_nl blk.9.ffn_gate_shexp.weight iq3_xxs blk.9.ffn_up_shexp.weight iq3_xxs blk.9.attn_k.weight iq3_xxs blk.9.attn_q.weight iq3_xxs blk.9.attn_v.weight q4_K +blk.10.ffn_down_exps.weight iq4_nl blk.10.ffn_gate_exps.weight iq3_xxs blk.10.ffn_up_exps.weight iq3_xxs +blk.10.ffn_down_shexp.weight iq4_nl blk.10.ffn_gate_shexp.weight iq3_xxs blk.10.ffn_up_shexp.weight iq3_xxs blk.10.attn_k.weight iq3_xxs blk.10.attn_q.weight iq3_xxs blk.10.attn_v.weight q4_K +blk.11.ffn_down_exps.weight iq4_nl blk.11.ffn_gate_exps.weight iq3_xxs blk.11.ffn_up_exps.weight iq3_xxs +blk.11.ffn_down_shexp.weight iq4_nl blk.11.ffn_gate_shexp.weight iq3_xxs blk.11.ffn_up_shexp.weight iq3_xxs blk.11.attn_k.weight iq3_xxs blk.11.attn_q.weight iq3_xxs blk.11.attn_v.weight q4_K +blk.12.ffn_down_exps.weight iq4_nl blk.12.ffn_gate_exps.weight iq3_xxs blk.12.ffn_up_exps.weight iq3_xxs +blk.12.ffn_down_shexp.weight iq4_nl blk.12.ffn_gate_shexp.weight iq3_xxs blk.12.ffn_up_shexp.weight iq3_xxs blk.12.attn_k.weight iq3_xxs blk.12.attn_q.weight iq3_xxs blk.12.attn_v.weight q4_K +blk.13.ffn_down_exps.weight iq4_nl blk.13.ffn_gate_exps.weight iq3_xxs blk.13.ffn_up_exps.weight iq3_xxs +blk.13.ffn_down_shexp.weight iq4_nl blk.13.ffn_gate_shexp.weight iq3_xxs blk.13.ffn_up_shexp.weight iq3_xxs blk.13.attn_k.weight iq3_xxs blk.13.attn_q.weight iq3_xxs blk.13.attn_v.weight q4_K +blk.14.ffn_down_exps.weight iq4_nl blk.14.ffn_gate_exps.weight iq3_xxs blk.14.ffn_up_exps.weight iq3_xxs +blk.14.ffn_down_shexp.weight iq4_nl blk.14.ffn_gate_shexp.weight iq3_xxs blk.14.ffn_up_shexp.weight iq3_xxs blk.14.attn_k.weight iq3_xxs blk.14.attn_q.weight iq3_xxs blk.14.attn_v.weight q4_K +blk.15.ffn_down_exps.weight iq4_nl blk.15.ffn_gate_exps.weight iq3_xxs blk.15.ffn_up_exps.weight iq3_xxs +blk.15.ffn_down_shexp.weight iq4_nl blk.15.ffn_gate_shexp.weight iq3_xxs blk.15.ffn_up_shexp.weight iq3_xxs blk.15.attn_k.weight iq3_xxs blk.15.attn_q.weight iq3_xxs blk.15.attn_v.weight q4_K +blk.16.ffn_down_exps.weight iq4_nl blk.16.ffn_gate_exps.weight iq3_xxs blk.16.ffn_up_exps.weight iq3_xxs +blk.16.ffn_down_shexp.weight iq4_nl blk.16.ffn_gate_shexp.weight iq3_xxs blk.16.ffn_up_shexp.weight iq3_xxs blk.16.attn_k.weight iq3_xxs blk.16.attn_q.weight iq3_xxs blk.16.attn_v.weight q4_K +blk.17.ffn_down_exps.weight iq4_nl blk.17.ffn_gate_exps.weight iq3_xxs blk.17.ffn_up_exps.weight iq3_xxs +blk.17.ffn_down_shexp.weight iq4_nl blk.17.ffn_gate_shexp.weight iq3_xxs blk.17.ffn_up_shexp.weight iq3_xxs blk.17.attn_k.weight iq3_xxs blk.17.attn_q.weight iq3_xxs blk.17.attn_v.weight q4_K +blk.18.ffn_down_exps.weight iq4_nl blk.18.ffn_gate_exps.weight iq3_xxs blk.18.ffn_up_exps.weight iq3_xxs +blk.18.ffn_down_shexp.weight iq4_nl blk.18.ffn_gate_shexp.weight iq3_xxs blk.18.ffn_up_shexp.weight iq3_xxs blk.18.attn_k.weight iq3_xxs blk.18.attn_q.weight iq3_xxs blk.18.attn_v.weight q4_K +blk.19.ffn_down_exps.weight iq4_nl blk.19.ffn_gate_exps.weight iq3_xxs blk.19.ffn_up_exps.weight iq3_xxs +blk.19.ffn_down_shexp.weight iq4_nl blk.19.ffn_gate_shexp.weight iq3_xxs blk.19.ffn_up_shexp.weight iq3_xxs blk.19.attn_k.weight iq3_xxs blk.19.attn_q.weight iq3_xxs blk.19.attn_v.weight q4_K +blk.20.ffn_down_exps.weight iq4_nl blk.20.ffn_gate_exps.weight iq3_xxs blk.20.ffn_up_exps.weight iq3_xxs +blk.20.ffn_down_shexp.weight iq4_nl blk.20.ffn_gate_shexp.weight iq3_xxs blk.20.ffn_up_shexp.weight iq3_xxs blk.20.attn_k.weight iq3_xxs blk.20.attn_q.weight iq3_xxs blk.20.attn_v.weight q4_K +blk.21.ffn_down_exps.weight iq4_nl blk.21.ffn_gate_exps.weight iq3_xxs blk.21.ffn_up_exps.weight iq3_xxs +blk.21.ffn_down_shexp.weight iq4_nl blk.21.ffn_gate_shexp.weight iq3_xxs blk.21.ffn_up_shexp.weight iq3_xxs blk.21.attn_k.weight iq3_xxs blk.21.attn_q.weight iq3_xxs blk.21.attn_v.weight q4_K +blk.22.ffn_down_exps.weight iq4_nl blk.22.ffn_gate_exps.weight iq3_xxs blk.22.ffn_up_exps.weight iq3_xxs +blk.22.ffn_down_shexp.weight iq4_nl blk.22.ffn_gate_shexp.weight iq3_xxs blk.22.ffn_up_shexp.weight iq3_xxs blk.22.attn_k.weight iq3_xxs blk.22.attn_q.weight iq3_xxs blk.22.attn_v.weight q4_K +blk.23.ffn_down_exps.weight iq4_nl blk.23.ffn_gate_exps.weight iq3_xxs blk.23.ffn_up_exps.weight iq3_xxs +blk.23.ffn_down_shexp.weight iq4_nl blk.23.ffn_gate_shexp.weight iq3_xxs blk.23.ffn_up_shexp.weight iq3_xxs blk.23.attn_k.weight iq3_xxs blk.23.attn_q.weight iq3_xxs blk.23.attn_v.weight q4_K +blk.24.ffn_down_exps.weight iq4_nl blk.24.ffn_gate_exps.weight iq3_xxs blk.24.ffn_up_exps.weight iq3_xxs +blk.24.ffn_down_shexp.weight iq4_nl blk.24.ffn_gate_shexp.weight iq3_xxs blk.24.ffn_up_shexp.weight iq3_xxs blk.24.attn_k.weight iq3_xxs blk.24.attn_q.weight iq3_xxs blk.24.attn_v.weight q4_K +blk.25.ffn_down_exps.weight iq4_nl blk.25.ffn_gate_exps.weight iq3_xxs blk.25.ffn_up_exps.weight iq3_xxs +blk.25.ffn_down_shexp.weight iq4_nl blk.25.ffn_gate_shexp.weight iq3_xxs blk.25.ffn_up_shexp.weight iq3_xxs blk.25.attn_k.weight iq3_xxs blk.25.attn_q.weight iq3_xxs blk.25.attn_v.weight q4_K +blk.26.ffn_down_exps.weight iq4_nl blk.26.ffn_gate_exps.weight iq3_xxs blk.26.ffn_up_exps.weight iq3_xxs +blk.26.ffn_down_shexp.weight iq4_nl blk.26.ffn_gate_shexp.weight iq3_xxs blk.26.ffn_up_shexp.weight iq3_xxs blk.26.attn_k.weight iq3_xxs blk.26.attn_q.weight iq3_xxs blk.26.attn_v.weight q4_K +blk.27.ffn_down_exps.weight iq4_nl blk.27.ffn_gate_exps.weight iq3_xxs blk.27.ffn_up_exps.weight iq3_xxs +blk.27.ffn_down_shexp.weight iq4_nl blk.27.ffn_gate_shexp.weight iq3_xxs blk.27.ffn_up_shexp.weight iq3_xxs blk.27.attn_k.weight iq3_xxs blk.27.attn_q.weight iq3_xxs blk.27.attn_v.weight q4_K +blk.28.ffn_down_exps.weight iq4_nl blk.28.ffn_gate_exps.weight iq3_xxs blk.28.ffn_up_exps.weight iq3_xxs +blk.28.ffn_down_shexp.weight iq4_nl blk.28.ffn_gate_shexp.weight iq3_xxs blk.28.ffn_up_shexp.weight iq3_xxs blk.28.attn_k.weight iq3_xxs blk.28.attn_q.weight iq3_xxs blk.28.attn_v.weight q4_K +blk.29.ffn_down_exps.weight iq4_nl blk.29.ffn_gate_exps.weight iq3_xxs blk.29.ffn_up_exps.weight iq3_xxs +blk.29.ffn_down_shexp.weight iq4_nl blk.29.ffn_gate_shexp.weight iq3_xxs blk.29.ffn_up_shexp.weight iq3_xxs blk.29.attn_k.weight iq3_xxs blk.29.attn_q.weight iq3_xxs blk.29.attn_v.weight q4_K +blk.30.ffn_down_exps.weight iq4_nl blk.30.ffn_gate_exps.weight iq3_xxs blk.30.ffn_up_exps.weight iq3_xxs +blk.30.ffn_down_shexp.weight iq4_nl blk.30.ffn_gate_shexp.weight iq3_xxs blk.30.ffn_up_shexp.weight iq3_xxs blk.30.attn_k.weight iq3_xxs blk.30.attn_q.weight iq3_xxs blk.30.attn_v.weight q4_K +blk.31.ffn_down_exps.weight iq4_nl blk.31.ffn_gate_exps.weight iq3_xxs blk.31.ffn_up_exps.weight iq3_xxs +blk.31.ffn_down_shexp.weight iq4_nl blk.31.ffn_gate_shexp.weight iq3_xxs blk.31.ffn_up_shexp.weight iq3_xxs blk.31.attn_k.weight iq3_xxs blk.31.attn_q.weight iq3_xxs blk.31.attn_v.weight q4_K +blk.32.ffn_down_exps.weight iq4_nl blk.32.ffn_gate_exps.weight iq3_xxs blk.32.ffn_up_exps.weight iq3_xxs +blk.32.ffn_down_shexp.weight iq4_nl blk.32.ffn_gate_shexp.weight iq3_xxs blk.32.ffn_up_shexp.weight iq3_xxs blk.32.attn_k.weight iq3_xxs blk.32.attn_q.weight iq3_xxs blk.32.attn_v.weight q4_K +blk.33.ffn_down_exps.weight iq4_nl blk.33.ffn_gate_exps.weight iq3_xxs blk.33.ffn_up_exps.weight iq3_xxs +blk.33.ffn_down_shexp.weight iq4_nl blk.33.ffn_gate_shexp.weight iq3_xxs blk.33.ffn_up_shexp.weight iq3_xxs blk.33.attn_k.weight iq3_xxs blk.33.attn_q.weight iq3_xxs blk.33.attn_v.weight q4_K +blk.34.ffn_down_exps.weight iq4_nl blk.34.ffn_gate_exps.weight iq3_xxs blk.34.ffn_up_exps.weight iq3_xxs blk.34.ffn_gate_shexp.weight iq3_xxs +blk.34.ffn_down_shexp.weight iq4_nl blk.34.ffn_up_shexp.weight iq3_xxs blk.34.attn_k.weight iq3_xxs blk.34.attn_q.weight iq3_xxs blk.34.attn_v.weight q4_K +blk.35.ffn_down_exps.weight iq4_nl blk.35.ffn_gate_exps.weight iq3_xxs blk.35.ffn_up_exps.weight iq3_xxs +blk.35.ffn_down_shexp.weight iq4_nl blk.35.ffn_gate_shexp.weight iq3_xxs blk.35.ffn_up_shexp.weight iq3_xxs blk.35.attn_k.weight iq3_xxs blk.35.attn_q.weight iq3_xxs blk.35.attn_v.weight q4_K +blk.36.ffn_down_exps.weight iq4_nl blk.36.ffn_gate_exps.weight iq3_xxs blk.36.ffn_up_exps.weight iq3_xxs +blk.36.ffn_down_shexp.weight iq4_nl blk.36.ffn_gate_shexp.weight iq3_xxs blk.36.ffn_up_shexp.weight iq3_xxs blk.36.attn_k.weight iq3_xxs blk.36.attn_q.weight iq3_xxs blk.36.attn_v.weight q4_K +blk.37.ffn_down_exps.weight iq4_nl blk.37.ffn_gate_exps.weight iq3_xxs blk.37.ffn_up_exps.weight iq3_xxs +blk.37.ffn_down_shexp.weight iq4_nl blk.37.ffn_gate_shexp.weight iq3_xxs blk.37.ffn_up_shexp.weight iq3_xxs blk.37.attn_k.weight iq3_xxs blk.37.attn_q.weight iq3_xxs blk.37.attn_v.weight q4_K +blk.38.ffn_down_exps.weight iq4_nl blk.38.ffn_gate_exps.weight iq3_xxs blk.38.ffn_up_exps.weight iq3_xxs +blk.38.ffn_down_shexp.weight iq4_nl blk.38.ffn_gate_shexp.weight iq3_xxs blk.38.ffn_up_shexp.weight iq3_xxs blk.38.attn_k.weight iq3_xxs blk.38.attn_q.weight iq3_xxs blk.38.attn_v.weight q4_K +blk.39.ffn_down_exps.weight iq4_nl blk.39.ffn_gate_exps.weight iq3_xxs blk.39.ffn_up_exps.weight iq3_xxs +blk.39.ffn_down_shexp.weight iq4_nl blk.39.ffn_gate_shexp.weight iq3_xxs blk.39.ffn_up_shexp.weight iq3_xxs blk.39.attn_k.weight iq3_xxs blk.39.attn_q.weight iq3_xxs blk.39.attn_v.weight q4_K +blk.40.ffn_down_exps.weight iq4_nl +blk.40.ffn_down_shexp.weight iq4_nl blk.40.attn_k.weight iq3_xxs blk.40.attn_q.weight iq3_xxs blk.40.attn_v.weight q4_K +blk.41.ffn_down_exps.weight iq4_nl +blk.41.ffn_down_shexp.weight iq4_nl blk.41.attn_k.weight iq3_xxs blk.41.attn_q.weight iq3_xxs blk.41.attn_v.weight q4_K +blk.42.ffn_down_exps.weight iq4_nl +blk.42.ffn_down_shexp.weight iq4_nl blk.42.attn_k.weight iq3_xxs blk.42.attn_q.weight iq3_xxs blk.42.attn_v.weight q4_K +blk.43.ffn_down_exps.weight iq4_nl +blk.43.ffn_down_shexp.weight iq4_nl blk.43.attn_k.weight iq3_xxs blk.43.attn_q.weight iq3_xxs blk.43.attn_v.weight q4_K +blk.44.ffn_down_exps.weight iq4_nl +blk.44.ffn_down_shexp.weight iq4_nl blk.44.attn_k.weight iq3_xxs blk.44.attn_q.weight iq3_xxs blk.44.attn_v.weight q4_K output.weight q6_K +blk.45.ffn_down_exps.weight iq4_nl +blk.45.ffn_down_shexp.weight iq4_nl blk.45.attn_k.weight iq3_xxs blk.45.attn_q.weight iq3_xxs blk.45.attn_v.weight q4_K [IQ3_XXS] iq3_xxs token_embd.weight iq3_s -blk.0.ffn_down.weight q4_K +blk.0.ffn_down.weight q5_0 blk.0.attn_k.weight iq2_s blk.0.attn_output.weight iq3_s blk.0.attn_q.weight iq2_s blk.0.attn_v.weight q4_K -blk.1.ffn_down_exps.weight q4_K -blk.1.ffn_down_shexp.weight q4_K +blk.1.ffn_down_exps.weight q5_0 +blk.1.ffn_down_shexp.weight q5_0 blk.1.attn_k.weight iq2_s blk.1.attn_output.weight iq3_s blk.1.attn_q.weight iq2_s blk.1.attn_v.weight q4_K -blk.2.ffn_down_exps.weight q4_K -blk.2.ffn_down_shexp.weight q4_K +blk.2.ffn_down_exps.weight q5_0 +blk.2.ffn_down_shexp.weight q5_0 blk.2.attn_k.weight iq2_s blk.2.attn_output.weight iq3_s blk.2.attn_q.weight iq2_s blk.2.attn_v.weight q4_K -blk.3.ffn_down_exps.weight q4_K -blk.3.ffn_down_shexp.weight q4_K +blk.3.ffn_down_exps.weight q5_0 +blk.3.ffn_down_shexp.weight q5_0 blk.3.attn_k.weight iq2_s blk.3.attn_output.weight iq3_s blk.3.attn_q.weight iq2_s blk.3.attn_v.weight q4_K -blk.4.ffn_down_exps.weight q4_K -blk.4.ffn_down_shexp.weight q4_K +blk.4.ffn_down_exps.weight q5_0 +blk.4.ffn_down_shexp.weight q5_0 blk.4.attn_k.weight iq2_s blk.4.attn_output.weight iq3_s blk.4.attn_q.weight iq2_s blk.4.attn_v.weight q4_K -blk.5.ffn_down_exps.weight q3_K -blk.5.ffn_down_shexp.weight q3_K +blk.5.ffn_down_exps.weight q4_0 +blk.5.ffn_down_shexp.weight q4_0 blk.5.attn_k.weight iq2_s blk.5.attn_output.weight iq3_s blk.5.attn_q.weight iq2_s blk.5.attn_v.weight q4_K -blk.6.ffn_down_exps.weight q3_K -blk.6.ffn_down_shexp.weight q3_K +blk.6.ffn_down_exps.weight q4_0 +blk.6.ffn_down_shexp.weight q4_0 blk.6.attn_k.weight iq2_s blk.6.attn_output.weight iq3_s blk.6.attn_q.weight iq2_s blk.6.attn_v.weight q4_K -blk.7.ffn_down_exps.weight q3_K -blk.7.ffn_down_shexp.weight q3_K +blk.7.ffn_down_exps.weight q4_0 +blk.7.ffn_down_shexp.weight q4_0 blk.7.attn_k.weight iq2_s blk.7.attn_output.weight iq3_s blk.7.attn_q.weight iq2_s blk.7.attn_v.weight q4_K -blk.8.ffn_down_exps.weight q3_K -blk.8.ffn_down_shexp.weight q3_K +blk.8.ffn_down_exps.weight q4_0 +blk.8.ffn_down_shexp.weight q4_0 blk.8.attn_k.weight iq2_s blk.8.attn_output.weight iq3_s blk.8.attn_q.weight iq2_s blk.8.attn_v.weight q4_K -blk.9.ffn_down_exps.weight q3_K -blk.9.ffn_down_shexp.weight q3_K +blk.9.ffn_down_exps.weight q4_0 +blk.9.ffn_down_shexp.weight q4_0 blk.9.attn_k.weight iq2_s blk.9.attn_output.weight iq3_s blk.9.attn_q.weight iq2_s blk.9.attn_v.weight q4_K -blk.10.ffn_down_exps.weight q3_K -blk.10.ffn_down_shexp.weight q3_K +blk.10.ffn_down_exps.weight q4_0 +blk.10.ffn_down_shexp.weight q4_0 blk.10.attn_k.weight iq2_s blk.10.attn_output.weight iq3_s blk.10.attn_q.weight iq2_s blk.10.attn_v.weight q4_K -blk.11.ffn_down_exps.weight q3_K -blk.11.ffn_down_shexp.weight q3_K +blk.11.ffn_down_exps.weight q4_0 +blk.11.ffn_down_shexp.weight q4_0 blk.11.attn_k.weight iq2_s blk.11.attn_output.weight iq3_s blk.11.attn_q.weight iq2_s blk.11.attn_v.weight q4_K -blk.12.ffn_down_exps.weight q3_K -blk.12.ffn_down_shexp.weight q3_K +blk.12.ffn_down_exps.weight q4_0 +blk.12.ffn_down_shexp.weight q4_0 blk.12.attn_k.weight iq2_s blk.12.attn_output.weight iq3_s blk.12.attn_q.weight iq2_s blk.12.attn_v.weight q4_K -blk.13.ffn_down_exps.weight q3_K -blk.13.ffn_down_shexp.weight q3_K +blk.13.ffn_down_exps.weight q4_0 +blk.13.ffn_down_shexp.weight q4_0 blk.13.attn_k.weight iq2_s blk.13.attn_output.weight iq3_s blk.13.attn_q.weight iq2_s blk.13.attn_v.weight q4_K -blk.14.ffn_down_exps.weight q3_K -blk.14.ffn_down_shexp.weight q3_K +blk.14.ffn_down_exps.weight q4_0 +blk.14.ffn_down_shexp.weight q4_0 blk.14.attn_k.weight iq2_s blk.14.attn_output.weight iq3_s blk.14.attn_q.weight iq2_s blk.14.attn_v.weight q4_K -blk.15.ffn_down_exps.weight q3_K -blk.15.ffn_down_shexp.weight q3_K +blk.15.ffn_down_exps.weight q4_0 +blk.15.ffn_down_shexp.weight q4_0 blk.15.attn_k.weight iq2_s blk.15.attn_output.weight iq3_s blk.15.attn_q.weight iq2_s blk.15.attn_v.weight q4_K -blk.16.ffn_down_exps.weight q3_K -blk.16.ffn_down_shexp.weight q3_K +blk.16.ffn_down_exps.weight q4_0 +blk.16.ffn_down_shexp.weight q4_0 blk.16.attn_k.weight iq2_s blk.16.attn_output.weight iq3_s blk.16.attn_q.weight iq2_s blk.16.attn_v.weight q4_K -blk.17.ffn_down_exps.weight q3_K -blk.17.ffn_down_shexp.weight q3_K +blk.17.ffn_down_exps.weight q4_0 +blk.17.ffn_down_shexp.weight q4_0 blk.17.attn_k.weight iq2_s blk.17.attn_output.weight iq3_s blk.17.attn_q.weight iq2_s blk.17.attn_v.weight q4_K -blk.18.ffn_down_exps.weight q3_K -blk.18.ffn_down_shexp.weight q3_K +blk.18.ffn_down_exps.weight q4_0 +blk.18.ffn_down_shexp.weight q4_0 blk.18.attn_k.weight iq2_s blk.18.attn_output.weight iq3_s blk.18.attn_q.weight iq2_s blk.18.attn_v.weight q4_K -blk.19.ffn_down_exps.weight q3_K -blk.19.ffn_down_shexp.weight q3_K +blk.19.ffn_down_exps.weight q4_0 +blk.19.ffn_down_shexp.weight q4_0 blk.19.attn_k.weight iq2_s blk.19.attn_output.weight iq3_s blk.19.attn_q.weight iq2_s blk.19.attn_v.weight q4_K -blk.20.ffn_down_exps.weight q3_K -blk.20.ffn_down_shexp.weight q3_K +blk.20.ffn_down_exps.weight q4_0 +blk.20.ffn_down_shexp.weight q4_0 blk.20.attn_k.weight iq2_s blk.20.attn_output.weight iq3_s blk.20.attn_q.weight iq2_s blk.20.attn_v.weight q4_K -blk.21.ffn_down_exps.weight q3_K -blk.21.ffn_down_shexp.weight q3_K +blk.21.ffn_down_exps.weight q4_0 +blk.21.ffn_down_shexp.weight q4_0 blk.21.attn_k.weight iq2_s blk.21.attn_output.weight iq3_s blk.21.attn_q.weight iq2_s blk.21.attn_v.weight q4_K -blk.22.ffn_down_exps.weight q3_K -blk.22.ffn_down_shexp.weight q3_K +blk.22.ffn_down_exps.weight q4_0 +blk.22.ffn_down_shexp.weight q4_0 blk.22.attn_k.weight iq2_s blk.22.attn_output.weight iq3_s blk.22.attn_q.weight iq2_s blk.22.attn_v.weight q4_K -blk.23.ffn_down_exps.weight q3_K -blk.23.ffn_down_shexp.weight q3_K +blk.23.ffn_down_exps.weight q4_0 +blk.23.ffn_down_shexp.weight q4_0 blk.23.attn_k.weight iq2_s blk.23.attn_output.weight iq3_s blk.23.attn_q.weight iq2_s blk.23.attn_v.weight q4_K -blk.24.ffn_down_exps.weight q3_K -blk.24.ffn_down_shexp.weight q3_K +blk.24.ffn_down_exps.weight q4_0 +blk.24.ffn_down_shexp.weight q4_0 blk.24.attn_k.weight iq2_s blk.24.attn_output.weight iq3_s blk.24.attn_q.weight iq2_s blk.24.attn_v.weight q4_K -blk.25.ffn_down_exps.weight q3_K -blk.25.ffn_down_shexp.weight q3_K +blk.25.ffn_down_exps.weight q4_0 +blk.25.ffn_down_shexp.weight q4_0 blk.25.attn_k.weight iq2_s blk.25.attn_output.weight iq3_s blk.25.attn_q.weight iq2_s blk.25.attn_v.weight q4_K -blk.26.ffn_down_exps.weight q3_K -blk.26.ffn_down_shexp.weight q3_K +blk.26.ffn_down_exps.weight q4_0 +blk.26.ffn_down_shexp.weight q4_0 blk.26.attn_k.weight iq2_s blk.26.attn_output.weight iq3_s blk.26.attn_q.weight iq2_s blk.26.attn_v.weight q4_K -blk.27.ffn_down_exps.weight q3_K -blk.27.ffn_down_shexp.weight q3_K +blk.27.ffn_down_exps.weight q4_0 +blk.27.ffn_down_shexp.weight q4_0 blk.27.attn_k.weight iq2_s blk.27.attn_output.weight iq3_s blk.27.attn_q.weight iq2_s blk.27.attn_v.weight q4_K -blk.28.ffn_down_exps.weight q3_K -blk.28.ffn_down_shexp.weight q3_K +blk.28.ffn_down_exps.weight q4_0 +blk.28.ffn_down_shexp.weight q4_0 blk.28.attn_k.weight iq2_s blk.28.attn_output.weight iq3_s blk.28.attn_q.weight iq2_s blk.28.attn_v.weight q4_K -blk.29.ffn_down_exps.weight q3_K -blk.29.ffn_down_shexp.weight q3_K +blk.29.ffn_down_exps.weight q4_0 +blk.29.ffn_down_shexp.weight q4_0 blk.29.attn_k.weight iq2_s blk.29.attn_output.weight iq3_s blk.29.attn_q.weight iq2_s blk.29.attn_v.weight q4_K -blk.30.ffn_down_exps.weight q3_K -blk.30.ffn_down_shexp.weight q3_K +blk.30.ffn_down_exps.weight q4_0 +blk.30.ffn_down_shexp.weight q4_0 blk.30.attn_k.weight iq2_s blk.30.attn_output.weight iq3_s blk.30.attn_q.weight iq2_s blk.30.attn_v.weight q4_K -blk.31.ffn_down_exps.weight q3_K -blk.31.ffn_down_shexp.weight q3_K +blk.31.ffn_down_exps.weight q4_0 +blk.31.ffn_down_shexp.weight q4_0 blk.31.attn_k.weight iq2_s blk.31.attn_output.weight iq3_s blk.31.attn_q.weight iq2_s blk.31.attn_v.weight q4_K -blk.32.ffn_down_exps.weight q3_K -blk.32.ffn_down_shexp.weight q3_K +blk.32.ffn_down_exps.weight q4_0 +blk.32.ffn_down_shexp.weight q4_0 blk.32.attn_k.weight iq2_s blk.32.attn_output.weight iq3_s blk.32.attn_q.weight iq2_s blk.32.attn_v.weight q4_K -blk.33.ffn_down_exps.weight q3_K -blk.33.ffn_down_shexp.weight q3_K +blk.33.ffn_down_exps.weight q4_0 +blk.33.ffn_down_shexp.weight q4_0 blk.33.attn_k.weight iq2_s blk.33.attn_output.weight iq3_s blk.33.attn_q.weight iq2_s blk.33.attn_v.weight q4_K -blk.34.ffn_down_exps.weight q3_K +blk.34.ffn_down_exps.weight q4_0 blk.34.attn_output.weight iq3_s -blk.34.ffn_down_shexp.weight q3_K +blk.34.ffn_down_shexp.weight q4_0 blk.34.attn_k.weight iq2_s blk.34.attn_q.weight iq2_s blk.34.attn_v.weight q4_K -blk.35.ffn_down_exps.weight q3_K -blk.35.ffn_down_shexp.weight q3_K +blk.35.ffn_down_exps.weight q4_0 +blk.35.ffn_down_shexp.weight q4_0 blk.35.attn_k.weight iq2_s blk.35.attn_output.weight iq3_s blk.35.attn_q.weight iq2_s blk.35.attn_v.weight q4_K -blk.36.ffn_down_exps.weight q3_K -blk.36.ffn_down_shexp.weight q3_K +blk.36.ffn_down_exps.weight q4_0 +blk.36.ffn_down_shexp.weight q4_0 blk.36.attn_k.weight iq2_s blk.36.attn_output.weight iq3_s blk.36.attn_q.weight iq2_s blk.36.attn_v.weight q4_K -blk.37.ffn_down_exps.weight q3_K -blk.37.ffn_down_shexp.weight q3_K +blk.37.ffn_down_exps.weight q4_0 +blk.37.ffn_down_shexp.weight q4_0 blk.37.attn_k.weight iq2_s blk.37.attn_output.weight iq3_s blk.37.attn_q.weight iq2_s blk.37.attn_v.weight q4_K -blk.38.ffn_down_exps.weight q3_K -blk.38.ffn_down_shexp.weight q3_K +blk.38.ffn_down_exps.weight q4_0 +blk.38.ffn_down_shexp.weight q4_0 blk.38.attn_k.weight iq2_s blk.38.attn_output.weight iq3_s blk.38.attn_q.weight iq2_s blk.38.attn_v.weight q4_K -blk.39.ffn_down_exps.weight q3_K -blk.39.ffn_down_shexp.weight q3_K +blk.39.ffn_down_exps.weight q4_0 +blk.39.ffn_down_shexp.weight q4_0 blk.39.attn_k.weight iq2_s blk.39.attn_output.weight iq3_s blk.39.attn_q.weight iq2_s blk.39.attn_v.weight q4_K -blk.40.ffn_down_exps.weight q3_K -blk.40.ffn_down_shexp.weight q3_K +blk.40.ffn_down_exps.weight q4_0 +blk.40.ffn_down_shexp.weight q4_0 blk.40.attn_k.weight iq2_s blk.40.attn_output.weight iq3_s blk.40.attn_q.weight iq2_s blk.40.attn_v.weight q4_K -blk.41.ffn_down_exps.weight q3_K -blk.41.ffn_down_shexp.weight q3_K +blk.41.ffn_down_exps.weight q4_0 +blk.41.ffn_down_shexp.weight q4_0 blk.41.attn_k.weight iq2_s blk.41.attn_output.weight iq3_s blk.41.attn_q.weight iq2_s blk.41.attn_v.weight q4_K -blk.42.ffn_down_exps.weight q3_K -blk.42.ffn_down_shexp.weight q3_K +blk.42.ffn_down_exps.weight q4_0 +blk.42.ffn_down_shexp.weight q4_0 blk.42.attn_k.weight iq2_s blk.42.attn_q.weight iq2_s blk.42.attn_v.weight q4_K blk.42.attn_output.weight iq3_s -blk.43.ffn_down_exps.weight q3_K -blk.43.ffn_down_shexp.weight q3_K +blk.43.ffn_down_exps.weight q4_0 +blk.43.ffn_down_shexp.weight q4_0 blk.43.attn_k.weight iq2_s blk.43.attn_output.weight iq3_s blk.43.attn_q.weight iq2_s blk.43.attn_v.weight q4_K -blk.44.ffn_down_exps.weight q3_K -blk.44.ffn_down_shexp.weight q3_K +blk.44.ffn_down_exps.weight q4_0 +blk.44.ffn_down_shexp.weight q4_0 blk.44.attn_k.weight iq2_s blk.44.attn_output.weight iq3_s blk.44.attn_q.weight iq2_s blk.44.attn_v.weight q4_K output.weight q5_K -blk.45.ffn_down_exps.weight q3_K -blk.45.ffn_down_shexp.weight q3_K +blk.45.ffn_down_exps.weight q4_0 +blk.45.ffn_down_shexp.weight q4_0 blk.45.attn_k.weight iq2_s blk.45.attn_output.weight iq3_s blk.45.attn_q.weight iq2_s @@ -1469,119 +2263,205 @@ blk.45.attn_v.weight q4_K [IQ1_S] iq1_s token_embd.weight q2_K -blk.0.ffn_down.weight q2_K +blk.0.ffn_down.weight q4_0 blk.0.attn_output.weight iq2_xxs blk.0.attn_v.weight q4_K -blk.1.ffn_down_exps.weight q2_K -blk.1.ffn_down_shexp.weight q2_K +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_down_shexp.weight q4_0 blk.1.attn_output.weight iq2_xxs blk.1.attn_v.weight q4_K -blk.2.ffn_down_exps.weight q2_K -blk.2.ffn_down_shexp.weight q2_K +blk.2.ffn_down_exps.weight q4_0 +blk.2.ffn_down_shexp.weight q4_0 blk.2.attn_output.weight iq2_xxs blk.2.attn_v.weight q4_K +blk.3.ffn_down_exps.weight iq4_nl +blk.3.ffn_down_shexp.weight iq4_nl blk.3.attn_output.weight iq2_xxs blk.3.attn_v.weight q4_K +blk.4.ffn_down_exps.weight iq4_nl +blk.4.ffn_down_shexp.weight iq4_nl blk.4.attn_output.weight iq2_xxs blk.4.attn_v.weight q4_K +blk.5.ffn_down_exps.weight iq4_nl +blk.5.ffn_down_shexp.weight iq4_nl blk.5.attn_output.weight iq2_xxs blk.5.attn_v.weight q4_K +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_down_shexp.weight iq4_nl blk.6.attn_output.weight iq2_xxs blk.6.attn_v.weight q4_K +blk.7.ffn_down_exps.weight iq4_nl +blk.7.ffn_down_shexp.weight iq4_nl blk.7.attn_output.weight iq2_xxs blk.7.attn_v.weight q4_K +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_down_shexp.weight iq4_nl blk.8.attn_output.weight iq2_xxs blk.8.attn_v.weight q4_K +blk.9.ffn_down_exps.weight iq4_nl +blk.9.ffn_down_shexp.weight iq4_nl blk.9.attn_output.weight iq2_xxs blk.9.attn_v.weight q4_K +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_down_shexp.weight iq4_nl blk.10.attn_output.weight iq2_xxs blk.10.attn_v.weight q4_K +blk.11.ffn_down_exps.weight iq4_nl +blk.11.ffn_down_shexp.weight iq4_nl blk.11.attn_output.weight iq2_xxs blk.11.attn_v.weight q4_K +blk.12.ffn_down_exps.weight iq4_nl +blk.12.ffn_down_shexp.weight iq4_nl blk.12.attn_output.weight iq2_xxs blk.12.attn_v.weight q4_K +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_down_shexp.weight iq4_nl blk.13.attn_output.weight iq2_xxs blk.13.attn_v.weight q4_K +blk.14.ffn_down_exps.weight iq4_nl +blk.14.ffn_down_shexp.weight iq4_nl blk.14.attn_output.weight iq2_xxs blk.14.attn_v.weight q4_K +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_down_shexp.weight iq4_nl blk.15.attn_output.weight iq2_xxs blk.15.attn_v.weight q4_K +blk.16.ffn_down_exps.weight iq4_nl +blk.16.ffn_down_shexp.weight iq4_nl blk.16.attn_output.weight iq2_xxs blk.16.attn_v.weight q4_K +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_down_shexp.weight iq4_nl blk.17.attn_output.weight iq2_xxs blk.17.attn_v.weight q4_K +blk.18.ffn_down_exps.weight iq4_nl +blk.18.ffn_down_shexp.weight iq4_nl blk.18.attn_output.weight iq2_xxs blk.18.attn_v.weight q4_K +blk.19.ffn_down_exps.weight iq4_nl +blk.19.ffn_down_shexp.weight iq4_nl blk.19.attn_output.weight iq2_xxs blk.19.attn_v.weight q4_K +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_down_shexp.weight iq4_nl blk.20.attn_output.weight iq2_xxs blk.20.attn_v.weight q4_K +blk.21.ffn_down_exps.weight iq4_nl +blk.21.ffn_down_shexp.weight iq4_nl blk.21.attn_output.weight iq2_xxs blk.21.attn_v.weight q4_K +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_down_shexp.weight iq4_nl blk.22.attn_output.weight iq2_xxs blk.22.attn_v.weight q4_K +blk.23.ffn_down_exps.weight iq4_nl +blk.23.ffn_down_shexp.weight iq4_nl blk.23.attn_output.weight iq2_xxs blk.23.attn_v.weight q4_K +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_down_shexp.weight iq4_nl blk.24.attn_output.weight iq2_xxs blk.24.attn_v.weight q4_K +blk.25.ffn_down_exps.weight iq4_nl +blk.25.ffn_down_shexp.weight iq4_nl blk.25.attn_output.weight iq2_xxs blk.25.attn_v.weight q4_K +blk.26.ffn_down_exps.weight iq4_nl +blk.26.ffn_down_shexp.weight iq4_nl blk.26.attn_output.weight iq2_xxs blk.26.attn_v.weight q4_K +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_down_shexp.weight iq4_nl blk.27.attn_output.weight iq2_xxs blk.27.attn_v.weight q4_K +blk.28.ffn_down_exps.weight iq4_nl +blk.28.ffn_down_shexp.weight iq4_nl blk.28.attn_output.weight iq2_xxs blk.28.attn_v.weight q4_K +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_down_shexp.weight iq4_nl blk.29.attn_output.weight iq2_xxs blk.29.attn_v.weight q4_K +blk.30.ffn_down_exps.weight iq4_nl +blk.30.ffn_down_shexp.weight iq4_nl blk.30.attn_output.weight iq2_xxs blk.30.attn_v.weight q4_K +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_down_shexp.weight iq4_nl blk.31.attn_output.weight iq2_xxs blk.31.attn_v.weight q4_K +blk.32.ffn_down_exps.weight iq4_nl +blk.32.ffn_down_shexp.weight iq4_nl blk.32.attn_output.weight iq2_xxs blk.32.attn_v.weight q4_K +blk.33.ffn_down_exps.weight iq4_nl +blk.33.ffn_down_shexp.weight iq4_nl blk.33.attn_output.weight iq2_xxs blk.33.attn_v.weight q4_K +blk.34.ffn_down_exps.weight iq4_nl blk.34.attn_output.weight iq2_xxs +blk.34.ffn_down_shexp.weight iq4_nl blk.34.attn_v.weight q4_K +blk.35.ffn_down_exps.weight iq4_nl +blk.35.ffn_down_shexp.weight iq4_nl blk.35.attn_output.weight iq2_xxs blk.35.attn_v.weight q4_K +blk.36.ffn_down_exps.weight iq4_nl +blk.36.ffn_down_shexp.weight iq4_nl blk.36.attn_output.weight iq2_xxs blk.36.attn_v.weight q4_K +blk.37.ffn_down_exps.weight iq4_nl +blk.37.ffn_down_shexp.weight iq4_nl blk.37.attn_output.weight iq2_xxs blk.37.attn_v.weight q4_K +blk.38.ffn_down_exps.weight iq4_nl +blk.38.ffn_down_shexp.weight iq4_nl blk.38.attn_output.weight iq2_xxs blk.38.attn_v.weight q4_K +blk.39.ffn_down_exps.weight iq4_nl +blk.39.ffn_down_shexp.weight iq4_nl blk.39.attn_output.weight iq2_xxs blk.39.attn_v.weight q4_K +blk.40.ffn_down_exps.weight iq4_nl +blk.40.ffn_down_shexp.weight iq4_nl blk.40.attn_output.weight iq2_xxs blk.40.attn_v.weight q4_K +blk.41.ffn_down_exps.weight iq4_nl +blk.41.ffn_down_shexp.weight iq4_nl blk.41.attn_output.weight iq2_xxs blk.41.attn_v.weight q4_K +blk.42.ffn_down_exps.weight iq4_nl +blk.42.ffn_down_shexp.weight iq4_nl blk.42.attn_v.weight q4_K blk.42.attn_output.weight iq2_xxs +blk.43.ffn_down_exps.weight iq4_nl +blk.43.ffn_down_shexp.weight iq4_nl blk.43.attn_output.weight iq2_xxs blk.43.attn_v.weight q4_K +blk.44.ffn_down_exps.weight iq4_nl +blk.44.ffn_down_shexp.weight iq4_nl blk.44.attn_output.weight iq2_xxs blk.44.attn_v.weight q4_K output.weight q5_K +blk.45.ffn_down_exps.weight iq4_nl +blk.45.ffn_down_shexp.weight iq4_nl blk.45.attn_output.weight iq2_xxs blk.45.attn_v.weight q4_K [IQ4_NL] iq4_nl -blk.0.ffn_down.weight q5_K +blk.0.ffn_down.weight q5_1 blk.0.attn_v.weight q5_K -blk.1.ffn_down_exps.weight q5_K -blk.1.ffn_down_shexp.weight q5_K +blk.1.ffn_down_exps.weight q5_1 +blk.1.ffn_down_shexp.weight q5_1 blk.1.attn_v.weight q5_K -blk.2.ffn_down_exps.weight q5_K -blk.2.ffn_down_shexp.weight q5_K +blk.2.ffn_down_exps.weight q5_1 +blk.2.ffn_down_shexp.weight q5_1 blk.2.attn_v.weight q5_K -blk.3.ffn_down_exps.weight q5_K -blk.3.ffn_down_shexp.weight q5_K +blk.3.ffn_down_exps.weight q5_1 +blk.3.ffn_down_shexp.weight q5_1 blk.3.attn_v.weight q5_K -blk.4.ffn_down_exps.weight q5_K -blk.4.ffn_down_shexp.weight q5_K +blk.4.ffn_down_exps.weight q5_1 +blk.4.ffn_down_shexp.weight q5_1 blk.4.attn_v.weight q5_K blk.5.attn_v.weight q5_K blk.6.attn_v.weight q5_K @@ -1627,529 +2507,1223 @@ output.weight q6_K blk.45.attn_v.weight q5_K [IQ3_S] iq3_s +blk.0.ffn_down.weight iq4_nl blk.0.attn_v.weight q4_K +blk.1.ffn_down_exps.weight iq4_nl +blk.1.ffn_down_shexp.weight iq4_nl blk.1.attn_v.weight q4_K +blk.2.ffn_down_exps.weight iq4_nl +blk.2.ffn_down_shexp.weight iq4_nl blk.2.attn_v.weight q4_K +blk.3.ffn_down_exps.weight iq4_nl +blk.3.ffn_down_shexp.weight iq4_nl blk.3.attn_v.weight q4_K +blk.4.ffn_down_exps.weight iq4_nl +blk.4.ffn_down_shexp.weight iq4_nl blk.4.attn_v.weight q4_K +blk.5.ffn_down_exps.weight iq4_nl +blk.5.ffn_down_shexp.weight iq4_nl blk.5.attn_v.weight q4_K +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_down_shexp.weight iq4_nl blk.6.attn_v.weight q4_K +blk.7.ffn_down_exps.weight iq4_nl +blk.7.ffn_down_shexp.weight iq4_nl blk.7.attn_v.weight q4_K +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_down_shexp.weight iq4_nl blk.8.attn_v.weight q4_K +blk.9.ffn_down_exps.weight iq4_nl +blk.9.ffn_down_shexp.weight iq4_nl blk.9.attn_v.weight q4_K +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_down_shexp.weight iq4_nl blk.10.attn_v.weight q4_K +blk.11.ffn_down_exps.weight iq4_nl +blk.11.ffn_down_shexp.weight iq4_nl blk.11.attn_v.weight q4_K +blk.12.ffn_down_exps.weight iq4_nl +blk.12.ffn_down_shexp.weight iq4_nl blk.12.attn_v.weight q4_K +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_down_shexp.weight iq4_nl blk.13.attn_v.weight q4_K +blk.14.ffn_down_exps.weight iq4_nl +blk.14.ffn_down_shexp.weight iq4_nl blk.14.attn_v.weight q4_K +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_down_shexp.weight iq4_nl blk.15.attn_v.weight q4_K +blk.16.ffn_down_exps.weight iq4_nl +blk.16.ffn_down_shexp.weight iq4_nl blk.16.attn_v.weight q4_K +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_down_shexp.weight iq4_nl blk.17.attn_v.weight q4_K +blk.18.ffn_down_exps.weight iq4_nl +blk.18.ffn_down_shexp.weight iq4_nl blk.18.attn_v.weight q4_K +blk.19.ffn_down_exps.weight iq4_nl +blk.19.ffn_down_shexp.weight iq4_nl blk.19.attn_v.weight q4_K +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_down_shexp.weight iq4_nl blk.20.attn_v.weight q4_K +blk.21.ffn_down_exps.weight iq4_nl +blk.21.ffn_down_shexp.weight iq4_nl blk.21.attn_v.weight q4_K +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_down_shexp.weight iq4_nl blk.22.attn_v.weight q4_K +blk.23.ffn_down_exps.weight iq4_nl +blk.23.ffn_down_shexp.weight iq4_nl blk.23.attn_v.weight q4_K +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_down_shexp.weight iq4_nl blk.24.attn_v.weight q4_K +blk.25.ffn_down_exps.weight iq4_nl +blk.25.ffn_down_shexp.weight iq4_nl blk.25.attn_v.weight q4_K +blk.26.ffn_down_exps.weight iq4_nl +blk.26.ffn_down_shexp.weight iq4_nl blk.26.attn_v.weight q4_K +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_down_shexp.weight iq4_nl blk.27.attn_v.weight q4_K +blk.28.ffn_down_exps.weight iq4_nl +blk.28.ffn_down_shexp.weight iq4_nl blk.28.attn_v.weight q4_K +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_down_shexp.weight iq4_nl blk.29.attn_v.weight q4_K +blk.30.ffn_down_exps.weight iq4_nl +blk.30.ffn_down_shexp.weight iq4_nl blk.30.attn_v.weight q4_K +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_down_shexp.weight iq4_nl blk.31.attn_v.weight q4_K +blk.32.ffn_down_exps.weight iq4_nl +blk.32.ffn_down_shexp.weight iq4_nl blk.32.attn_v.weight q4_K +blk.33.ffn_down_exps.weight iq4_nl +blk.33.ffn_down_shexp.weight iq4_nl blk.33.attn_v.weight q4_K +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_down_shexp.weight iq4_nl blk.34.attn_v.weight q4_K +blk.35.ffn_down_exps.weight iq4_nl +blk.35.ffn_down_shexp.weight iq4_nl blk.35.attn_v.weight q4_K +blk.36.ffn_down_exps.weight iq4_nl +blk.36.ffn_down_shexp.weight iq4_nl blk.36.attn_v.weight q4_K +blk.37.ffn_down_exps.weight iq4_nl +blk.37.ffn_down_shexp.weight iq4_nl blk.37.attn_v.weight q4_K +blk.38.ffn_down_exps.weight iq4_nl +blk.38.ffn_down_shexp.weight iq4_nl blk.38.attn_v.weight q4_K +blk.39.ffn_down_exps.weight iq4_nl +blk.39.ffn_down_shexp.weight iq4_nl blk.39.attn_v.weight q4_K +blk.40.ffn_down_exps.weight iq4_nl +blk.40.ffn_down_shexp.weight iq4_nl blk.40.attn_v.weight q4_K +blk.41.ffn_down_exps.weight iq4_nl +blk.41.ffn_down_shexp.weight iq4_nl blk.41.attn_v.weight q4_K +blk.42.ffn_down_exps.weight iq4_nl +blk.42.ffn_down_shexp.weight iq4_nl blk.42.attn_v.weight q4_K +blk.43.ffn_down_exps.weight iq4_nl +blk.43.ffn_down_shexp.weight iq4_nl blk.43.attn_v.weight q4_K +blk.44.ffn_down_exps.weight iq4_nl +blk.44.ffn_down_shexp.weight iq4_nl blk.44.attn_v.weight q4_K output.weight q6_K +blk.45.ffn_down_exps.weight iq4_nl +blk.45.ffn_down_shexp.weight iq4_nl blk.45.attn_v.weight q4_K [IQ3_M] iq3_s -blk.0.ffn_down.weight q4_K +blk.0.ffn_down.weight q5_0 blk.0.attn_output.weight q4_K blk.0.attn_v.weight q4_K -blk.1.ffn_down_exps.weight q4_K -blk.1.ffn_down_shexp.weight q4_K +blk.1.ffn_down_exps.weight q5_0 +blk.1.ffn_down_shexp.weight q5_0 blk.1.attn_output.weight q4_K blk.1.attn_v.weight q4_K -blk.2.ffn_down_exps.weight q4_K -blk.2.ffn_down_shexp.weight q4_K +blk.2.ffn_down_exps.weight q5_0 +blk.2.ffn_down_shexp.weight q5_0 blk.2.attn_output.weight q4_K blk.2.attn_v.weight q4_K -blk.3.ffn_down_exps.weight q4_K -blk.3.ffn_down_shexp.weight q4_K +blk.3.ffn_down_exps.weight q5_0 +blk.3.ffn_down_shexp.weight q5_0 blk.3.attn_output.weight q4_K blk.3.attn_v.weight q4_K -blk.4.ffn_down_exps.weight q4_K -blk.4.ffn_down_shexp.weight q4_K +blk.4.ffn_down_exps.weight q5_0 +blk.4.ffn_down_shexp.weight q5_0 blk.4.attn_output.weight q4_K blk.4.attn_v.weight q4_K +blk.5.ffn_down_exps.weight iq4_nl +blk.5.ffn_down_shexp.weight iq4_nl blk.5.attn_output.weight q4_K blk.5.attn_v.weight q4_K +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_down_shexp.weight iq4_nl blk.6.attn_output.weight q4_K blk.6.attn_v.weight q4_K +blk.7.ffn_down_exps.weight iq4_nl +blk.7.ffn_down_shexp.weight iq4_nl blk.7.attn_output.weight q4_K blk.7.attn_v.weight q4_K +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_down_shexp.weight iq4_nl blk.8.attn_output.weight q4_K blk.8.attn_v.weight q4_K +blk.9.ffn_down_exps.weight iq4_nl +blk.9.ffn_down_shexp.weight iq4_nl blk.9.attn_output.weight q4_K blk.9.attn_v.weight q4_K +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_down_shexp.weight iq4_nl blk.10.attn_output.weight q4_K blk.10.attn_v.weight q4_K +blk.11.ffn_down_exps.weight iq4_nl +blk.11.ffn_down_shexp.weight iq4_nl blk.11.attn_output.weight q4_K blk.11.attn_v.weight q4_K +blk.12.ffn_down_exps.weight iq4_nl +blk.12.ffn_down_shexp.weight iq4_nl blk.12.attn_output.weight q4_K blk.12.attn_v.weight q4_K +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_down_shexp.weight iq4_nl blk.13.attn_output.weight q4_K blk.13.attn_v.weight q4_K +blk.14.ffn_down_exps.weight iq4_nl +blk.14.ffn_down_shexp.weight iq4_nl blk.14.attn_output.weight q4_K blk.14.attn_v.weight q4_K +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_down_shexp.weight iq4_nl blk.15.attn_output.weight q4_K blk.15.attn_v.weight q4_K +blk.16.ffn_down_exps.weight iq4_nl +blk.16.ffn_down_shexp.weight iq4_nl blk.16.attn_output.weight q4_K blk.16.attn_v.weight q4_K +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_down_shexp.weight iq4_nl blk.17.attn_output.weight q4_K blk.17.attn_v.weight q4_K +blk.18.ffn_down_exps.weight iq4_nl +blk.18.ffn_down_shexp.weight iq4_nl blk.18.attn_output.weight q4_K blk.18.attn_v.weight q4_K +blk.19.ffn_down_exps.weight iq4_nl +blk.19.ffn_down_shexp.weight iq4_nl blk.19.attn_output.weight q4_K blk.19.attn_v.weight q4_K +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_down_shexp.weight iq4_nl blk.20.attn_output.weight q4_K blk.20.attn_v.weight q4_K +blk.21.ffn_down_exps.weight iq4_nl +blk.21.ffn_down_shexp.weight iq4_nl blk.21.attn_output.weight q4_K blk.21.attn_v.weight q4_K +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_down_shexp.weight iq4_nl blk.22.attn_output.weight q4_K blk.22.attn_v.weight q4_K +blk.23.ffn_down_exps.weight iq4_nl +blk.23.ffn_down_shexp.weight iq4_nl blk.23.attn_output.weight q4_K blk.23.attn_v.weight q4_K +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_down_shexp.weight iq4_nl blk.24.attn_output.weight q4_K blk.24.attn_v.weight q4_K +blk.25.ffn_down_exps.weight iq4_nl +blk.25.ffn_down_shexp.weight iq4_nl blk.25.attn_output.weight q4_K blk.25.attn_v.weight q4_K +blk.26.ffn_down_exps.weight iq4_nl +blk.26.ffn_down_shexp.weight iq4_nl blk.26.attn_output.weight q4_K blk.26.attn_v.weight q4_K +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_down_shexp.weight iq4_nl blk.27.attn_output.weight q4_K blk.27.attn_v.weight q4_K +blk.28.ffn_down_exps.weight iq4_nl +blk.28.ffn_down_shexp.weight iq4_nl blk.28.attn_output.weight q4_K blk.28.attn_v.weight q4_K +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_down_shexp.weight iq4_nl blk.29.attn_output.weight q4_K blk.29.attn_v.weight q4_K +blk.30.ffn_down_exps.weight iq4_nl +blk.30.ffn_down_shexp.weight iq4_nl blk.30.attn_output.weight q4_K blk.30.attn_v.weight q4_K +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_down_shexp.weight iq4_nl blk.31.attn_output.weight q4_K blk.31.attn_v.weight q4_K +blk.32.ffn_down_exps.weight iq4_nl +blk.32.ffn_down_shexp.weight iq4_nl blk.32.attn_output.weight q4_K blk.32.attn_v.weight q4_K +blk.33.ffn_down_exps.weight iq4_nl +blk.33.ffn_down_shexp.weight iq4_nl blk.33.attn_output.weight q4_K blk.33.attn_v.weight q4_K +blk.34.ffn_down_exps.weight iq4_nl blk.34.attn_output.weight q4_K +blk.34.ffn_down_shexp.weight iq4_nl blk.34.attn_v.weight q4_K +blk.35.ffn_down_exps.weight iq4_nl +blk.35.ffn_down_shexp.weight iq4_nl blk.35.attn_output.weight q4_K blk.35.attn_v.weight q4_K +blk.36.ffn_down_exps.weight iq4_nl +blk.36.ffn_down_shexp.weight iq4_nl blk.36.attn_output.weight q4_K blk.36.attn_v.weight q4_K +blk.37.ffn_down_exps.weight iq4_nl +blk.37.ffn_down_shexp.weight iq4_nl blk.37.attn_output.weight q4_K blk.37.attn_v.weight q4_K +blk.38.ffn_down_exps.weight iq4_nl +blk.38.ffn_down_shexp.weight iq4_nl blk.38.attn_output.weight q4_K blk.38.attn_v.weight q4_K +blk.39.ffn_down_exps.weight iq4_nl +blk.39.ffn_down_shexp.weight iq4_nl blk.39.attn_output.weight q4_K blk.39.attn_v.weight q4_K +blk.40.ffn_down_exps.weight iq4_nl +blk.40.ffn_down_shexp.weight iq4_nl blk.40.attn_output.weight q4_K blk.40.attn_v.weight q4_K +blk.41.ffn_down_exps.weight iq4_nl +blk.41.ffn_down_shexp.weight iq4_nl blk.41.attn_output.weight q4_K blk.41.attn_v.weight q4_K +blk.42.ffn_down_exps.weight iq4_nl +blk.42.ffn_down_shexp.weight iq4_nl blk.42.attn_v.weight q4_K blk.42.attn_output.weight q4_K +blk.43.ffn_down_exps.weight iq4_nl +blk.43.ffn_down_shexp.weight iq4_nl blk.43.attn_output.weight q4_K blk.43.attn_v.weight q4_K +blk.44.ffn_down_exps.weight iq4_nl +blk.44.ffn_down_shexp.weight iq4_nl blk.44.attn_output.weight q4_K blk.44.attn_v.weight q4_K output.weight q6_K +blk.45.ffn_down_exps.weight iq4_nl +blk.45.ffn_down_shexp.weight iq4_nl blk.45.attn_output.weight q4_K blk.45.attn_v.weight q4_K [IQ2_S] iq2_xs token_embd.weight iq3_s -blk.0.ffn_down.weight iq3_s +blk.0.ffn_down.weight iq4_nl blk.0.attn_output.weight iq3_s blk.0.attn_v.weight q4_K -blk.1.ffn_down_exps.weight iq3_s -blk.1.ffn_down_shexp.weight iq3_s +blk.1.ffn_down_exps.weight iq4_nl +blk.1.ffn_down_shexp.weight iq4_nl blk.1.attn_output.weight iq3_s blk.1.attn_v.weight q4_K -blk.2.ffn_down_exps.weight iq3_s -blk.2.ffn_down_shexp.weight iq3_s +blk.2.ffn_down_exps.weight iq4_nl +blk.2.ffn_down_shexp.weight iq4_nl blk.2.attn_output.weight iq3_s blk.2.attn_v.weight q4_K +blk.3.ffn_down_exps.weight iq4_nl +blk.3.ffn_down_shexp.weight iq4_nl blk.3.attn_output.weight iq3_s blk.3.attn_v.weight q4_K +blk.4.ffn_down_exps.weight iq4_nl +blk.4.ffn_down_shexp.weight iq4_nl blk.4.attn_output.weight iq3_s blk.4.attn_v.weight q4_K +blk.5.ffn_down_exps.weight iq4_nl +blk.5.ffn_down_shexp.weight iq4_nl blk.5.attn_output.weight iq3_s blk.5.attn_v.weight q4_K +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_down_shexp.weight iq4_nl blk.6.attn_output.weight iq3_s blk.6.attn_v.weight q4_K +blk.7.ffn_down_exps.weight iq4_nl +blk.7.ffn_down_shexp.weight iq4_nl blk.7.attn_output.weight iq3_s blk.7.attn_v.weight q4_K +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_down_shexp.weight iq4_nl blk.8.attn_output.weight iq3_s blk.8.attn_v.weight q4_K +blk.9.ffn_down_exps.weight iq4_nl +blk.9.ffn_down_shexp.weight iq4_nl blk.9.attn_output.weight iq3_s blk.9.attn_v.weight q4_K +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_down_shexp.weight iq4_nl blk.10.attn_output.weight iq3_s blk.10.attn_v.weight q4_K +blk.11.ffn_down_exps.weight iq4_nl +blk.11.ffn_down_shexp.weight iq4_nl blk.11.attn_output.weight iq3_s blk.11.attn_v.weight q4_K +blk.12.ffn_down_exps.weight iq4_nl +blk.12.ffn_down_shexp.weight iq4_nl blk.12.attn_output.weight iq3_s blk.12.attn_v.weight q4_K +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_down_shexp.weight iq4_nl blk.13.attn_output.weight iq3_s blk.13.attn_v.weight q4_K +blk.14.ffn_down_exps.weight iq4_nl +blk.14.ffn_down_shexp.weight iq4_nl blk.14.attn_output.weight iq3_s blk.14.attn_v.weight q4_K +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_down_shexp.weight iq4_nl blk.15.attn_output.weight iq3_s blk.15.attn_v.weight q4_K +blk.16.ffn_down_exps.weight iq4_nl +blk.16.ffn_down_shexp.weight iq4_nl blk.16.attn_output.weight iq3_s blk.16.attn_v.weight q4_K +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_down_shexp.weight iq4_nl blk.17.attn_output.weight iq3_s blk.17.attn_v.weight q4_K +blk.18.ffn_down_exps.weight iq4_nl +blk.18.ffn_down_shexp.weight iq4_nl blk.18.attn_output.weight iq3_s blk.18.attn_v.weight q4_K +blk.19.ffn_down_exps.weight iq4_nl +blk.19.ffn_down_shexp.weight iq4_nl blk.19.attn_output.weight iq3_s blk.19.attn_v.weight q4_K +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_down_shexp.weight iq4_nl blk.20.attn_output.weight iq3_s blk.20.attn_v.weight q4_K +blk.21.ffn_down_exps.weight iq4_nl +blk.21.ffn_down_shexp.weight iq4_nl blk.21.attn_output.weight iq3_s blk.21.attn_v.weight q4_K +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_down_shexp.weight iq4_nl blk.22.attn_output.weight iq3_s blk.22.attn_v.weight q4_K +blk.23.ffn_down_exps.weight iq4_nl +blk.23.ffn_down_shexp.weight iq4_nl blk.23.attn_output.weight iq3_s blk.23.attn_v.weight q4_K +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_down_shexp.weight iq4_nl blk.24.attn_output.weight iq3_s blk.24.attn_v.weight q4_K +blk.25.ffn_down_exps.weight iq4_nl +blk.25.ffn_down_shexp.weight iq4_nl blk.25.attn_output.weight iq3_s blk.25.attn_v.weight q4_K +blk.26.ffn_down_exps.weight iq4_nl +blk.26.ffn_down_shexp.weight iq4_nl blk.26.attn_output.weight iq3_s blk.26.attn_v.weight q4_K +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_down_shexp.weight iq4_nl blk.27.attn_output.weight iq3_s blk.27.attn_v.weight q4_K +blk.28.ffn_down_exps.weight iq4_nl +blk.28.ffn_down_shexp.weight iq4_nl blk.28.attn_output.weight iq3_s blk.28.attn_v.weight q4_K +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_down_shexp.weight iq4_nl blk.29.attn_output.weight iq3_s blk.29.attn_v.weight q4_K +blk.30.ffn_down_exps.weight iq4_nl +blk.30.ffn_down_shexp.weight iq4_nl blk.30.attn_output.weight iq3_s blk.30.attn_v.weight q4_K +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_down_shexp.weight iq4_nl blk.31.attn_output.weight iq3_s blk.31.attn_v.weight q4_K +blk.32.ffn_down_exps.weight iq4_nl +blk.32.ffn_down_shexp.weight iq4_nl blk.32.attn_output.weight iq3_s blk.32.attn_v.weight q4_K +blk.33.ffn_down_exps.weight iq4_nl +blk.33.ffn_down_shexp.weight iq4_nl blk.33.attn_output.weight iq3_s blk.33.attn_v.weight q4_K +blk.34.ffn_down_exps.weight iq4_nl blk.34.attn_output.weight iq3_s +blk.34.ffn_down_shexp.weight iq4_nl blk.34.attn_v.weight q4_K +blk.35.ffn_down_exps.weight iq4_nl +blk.35.ffn_down_shexp.weight iq4_nl blk.35.attn_output.weight iq3_s blk.35.attn_v.weight q4_K +blk.36.ffn_down_exps.weight iq4_nl +blk.36.ffn_down_shexp.weight iq4_nl blk.36.attn_output.weight iq3_s blk.36.attn_v.weight q4_K +blk.37.ffn_down_exps.weight iq4_nl +blk.37.ffn_down_shexp.weight iq4_nl blk.37.attn_output.weight iq3_s blk.37.attn_v.weight q4_K +blk.38.ffn_down_exps.weight iq4_nl +blk.38.ffn_down_shexp.weight iq4_nl blk.38.attn_output.weight iq3_s blk.38.attn_v.weight q4_K +blk.39.ffn_down_exps.weight iq4_nl +blk.39.ffn_down_shexp.weight iq4_nl blk.39.attn_output.weight iq3_s blk.39.attn_v.weight q4_K +blk.40.ffn_down_exps.weight iq4_nl +blk.40.ffn_down_shexp.weight iq4_nl blk.40.attn_output.weight iq3_s blk.40.attn_v.weight q4_K +blk.41.ffn_down_exps.weight iq4_nl +blk.41.ffn_down_shexp.weight iq4_nl blk.41.attn_output.weight iq3_s blk.41.attn_v.weight q4_K +blk.42.ffn_down_exps.weight iq4_nl +blk.42.ffn_down_shexp.weight iq4_nl blk.42.attn_v.weight q4_K blk.42.attn_output.weight iq3_s +blk.43.ffn_down_exps.weight iq4_nl +blk.43.ffn_down_shexp.weight iq4_nl blk.43.attn_output.weight iq3_s blk.43.attn_v.weight q4_K +blk.44.ffn_down_exps.weight iq4_nl +blk.44.ffn_down_shexp.weight iq4_nl blk.44.attn_output.weight iq3_s blk.44.attn_v.weight q4_K output.weight q5_K +blk.45.ffn_down_exps.weight iq4_nl +blk.45.ffn_down_shexp.weight iq4_nl blk.45.attn_output.weight iq3_s blk.45.attn_v.weight q4_K [IQ2_M] iq2_s token_embd.weight iq3_s -blk.0.ffn_down.weight iq3_s +blk.0.ffn_down.weight iq4_nl blk.0.attn_output.weight iq3_s blk.0.attn_v.weight q4_K -blk.1.ffn_down_exps.weight iq3_s -blk.1.ffn_down_shexp.weight iq3_s +blk.1.ffn_down_exps.weight iq4_nl +blk.1.ffn_down_shexp.weight iq4_nl blk.1.attn_output.weight iq3_s blk.1.attn_v.weight q4_K -blk.2.ffn_down_exps.weight iq3_s -blk.2.ffn_down_shexp.weight iq3_s +blk.2.ffn_down_exps.weight iq4_nl +blk.2.ffn_down_shexp.weight iq4_nl blk.2.attn_output.weight iq3_s blk.2.attn_v.weight q4_K +blk.3.ffn_down_exps.weight iq4_nl +blk.3.ffn_down_shexp.weight iq4_nl blk.3.attn_output.weight iq3_s blk.3.attn_v.weight q4_K +blk.4.ffn_down_exps.weight iq4_nl +blk.4.ffn_down_shexp.weight iq4_nl blk.4.attn_output.weight iq3_s blk.4.attn_v.weight q4_K +blk.5.ffn_down_exps.weight iq4_nl +blk.5.ffn_down_shexp.weight iq4_nl blk.5.attn_output.weight iq3_s blk.5.attn_v.weight q4_K +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_down_shexp.weight iq4_nl blk.6.attn_output.weight iq3_s blk.6.attn_v.weight q4_K +blk.7.ffn_down_exps.weight iq4_nl +blk.7.ffn_down_shexp.weight iq4_nl blk.7.attn_output.weight iq3_s blk.7.attn_v.weight q4_K +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_down_shexp.weight iq4_nl blk.8.attn_output.weight iq3_s blk.8.attn_v.weight q4_K +blk.9.ffn_down_exps.weight iq4_nl +blk.9.ffn_down_shexp.weight iq4_nl blk.9.attn_output.weight iq3_s blk.9.attn_v.weight q4_K +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_down_shexp.weight iq4_nl blk.10.attn_output.weight iq3_s blk.10.attn_v.weight q4_K +blk.11.ffn_down_exps.weight iq4_nl +blk.11.ffn_down_shexp.weight iq4_nl blk.11.attn_output.weight iq3_s blk.11.attn_v.weight q4_K +blk.12.ffn_down_exps.weight iq4_nl +blk.12.ffn_down_shexp.weight iq4_nl blk.12.attn_output.weight iq3_s blk.12.attn_v.weight q4_K +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_down_shexp.weight iq4_nl blk.13.attn_output.weight iq3_s blk.13.attn_v.weight q4_K +blk.14.ffn_down_exps.weight iq4_nl +blk.14.ffn_down_shexp.weight iq4_nl blk.14.attn_output.weight iq3_s blk.14.attn_v.weight q4_K +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_down_shexp.weight iq4_nl blk.15.attn_output.weight iq3_s blk.15.attn_v.weight q4_K +blk.16.ffn_down_exps.weight iq4_nl +blk.16.ffn_down_shexp.weight iq4_nl blk.16.attn_output.weight iq3_s blk.16.attn_v.weight q4_K +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_down_shexp.weight iq4_nl blk.17.attn_output.weight iq3_s blk.17.attn_v.weight q4_K +blk.18.ffn_down_exps.weight iq4_nl +blk.18.ffn_down_shexp.weight iq4_nl blk.18.attn_output.weight iq3_s blk.18.attn_v.weight q4_K +blk.19.ffn_down_exps.weight iq4_nl +blk.19.ffn_down_shexp.weight iq4_nl blk.19.attn_output.weight iq3_s blk.19.attn_v.weight q4_K +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_down_shexp.weight iq4_nl blk.20.attn_output.weight iq3_s blk.20.attn_v.weight q4_K +blk.21.ffn_down_exps.weight iq4_nl +blk.21.ffn_down_shexp.weight iq4_nl blk.21.attn_output.weight iq3_s blk.21.attn_v.weight q4_K +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_down_shexp.weight iq4_nl blk.22.attn_output.weight iq3_s blk.22.attn_v.weight q4_K +blk.23.ffn_down_exps.weight iq4_nl +blk.23.ffn_down_shexp.weight iq4_nl blk.23.attn_output.weight iq3_s blk.23.attn_v.weight q4_K +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_down_shexp.weight iq4_nl blk.24.attn_output.weight iq3_s blk.24.attn_v.weight q4_K +blk.25.ffn_down_exps.weight iq4_nl +blk.25.ffn_down_shexp.weight iq4_nl blk.25.attn_output.weight iq3_s blk.25.attn_v.weight q4_K +blk.26.ffn_down_exps.weight iq4_nl +blk.26.ffn_down_shexp.weight iq4_nl blk.26.attn_output.weight iq3_s blk.26.attn_v.weight q4_K +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_down_shexp.weight iq4_nl blk.27.attn_output.weight iq3_s blk.27.attn_v.weight q4_K +blk.28.ffn_down_exps.weight iq4_nl +blk.28.ffn_down_shexp.weight iq4_nl blk.28.attn_output.weight iq3_s blk.28.attn_v.weight q4_K +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_down_shexp.weight iq4_nl blk.29.attn_output.weight iq3_s blk.29.attn_v.weight q4_K +blk.30.ffn_down_exps.weight iq4_nl +blk.30.ffn_down_shexp.weight iq4_nl blk.30.attn_output.weight iq3_s blk.30.attn_v.weight q4_K +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_down_shexp.weight iq4_nl blk.31.attn_output.weight iq3_s blk.31.attn_v.weight q4_K +blk.32.ffn_down_exps.weight iq4_nl +blk.32.ffn_down_shexp.weight iq4_nl blk.32.attn_output.weight iq3_s blk.32.attn_v.weight q4_K +blk.33.ffn_down_exps.weight iq4_nl +blk.33.ffn_down_shexp.weight iq4_nl blk.33.attn_output.weight iq3_s blk.33.attn_v.weight q4_K +blk.34.ffn_down_exps.weight iq4_nl blk.34.attn_output.weight iq3_s +blk.34.ffn_down_shexp.weight iq4_nl blk.34.attn_v.weight q4_K +blk.35.ffn_down_exps.weight iq4_nl +blk.35.ffn_down_shexp.weight iq4_nl blk.35.attn_output.weight iq3_s blk.35.attn_v.weight q4_K +blk.36.ffn_down_exps.weight iq4_nl +blk.36.ffn_down_shexp.weight iq4_nl blk.36.attn_output.weight iq3_s blk.36.attn_v.weight q4_K +blk.37.ffn_down_exps.weight iq4_nl +blk.37.ffn_down_shexp.weight iq4_nl blk.37.attn_output.weight iq3_s blk.37.attn_v.weight q4_K +blk.38.ffn_down_exps.weight iq4_nl +blk.38.ffn_down_shexp.weight iq4_nl blk.38.attn_output.weight iq3_s blk.38.attn_v.weight q4_K +blk.39.ffn_down_exps.weight iq4_nl +blk.39.ffn_down_shexp.weight iq4_nl blk.39.attn_output.weight iq3_s blk.39.attn_v.weight q4_K +blk.40.ffn_down_exps.weight iq4_nl +blk.40.ffn_down_shexp.weight iq4_nl blk.40.attn_output.weight iq3_s blk.40.attn_v.weight q4_K +blk.41.ffn_down_exps.weight iq4_nl +blk.41.ffn_down_shexp.weight iq4_nl blk.41.attn_output.weight iq3_s blk.41.attn_v.weight q4_K +blk.42.ffn_down_exps.weight iq4_nl +blk.42.ffn_down_shexp.weight iq4_nl blk.42.attn_v.weight q4_K blk.42.attn_output.weight iq3_s +blk.43.ffn_down_exps.weight iq4_nl +blk.43.ffn_down_shexp.weight iq4_nl blk.43.attn_output.weight iq3_s blk.43.attn_v.weight q4_K +blk.44.ffn_down_exps.weight iq4_nl +blk.44.ffn_down_shexp.weight iq4_nl blk.44.attn_output.weight iq3_s blk.44.attn_v.weight q4_K output.weight q5_K +blk.45.ffn_down_exps.weight iq4_nl +blk.45.ffn_down_shexp.weight iq4_nl blk.45.attn_output.weight iq3_s blk.45.attn_v.weight q4_K [IQ4_XS] iq4_xs -blk.0.ffn_down.weight q5_K +blk.0.ffn_down.weight q5_1 blk.0.attn_v.weight q5_K -blk.1.ffn_down_exps.weight q5_K -blk.1.ffn_down_shexp.weight q5_K +blk.1.ffn_down_exps.weight q5_1 +blk.1.ffn_down_shexp.weight q5_1 blk.1.attn_v.weight q5_K -blk.2.ffn_down_exps.weight q5_K -blk.2.ffn_down_shexp.weight q5_K +blk.2.ffn_down_exps.weight q5_1 +blk.2.ffn_down_shexp.weight q5_1 blk.2.attn_v.weight q5_K -blk.3.ffn_down_exps.weight q5_K -blk.3.ffn_down_shexp.weight q5_K +blk.3.ffn_down_exps.weight q5_1 +blk.3.ffn_down_shexp.weight q5_1 blk.3.attn_v.weight q5_K -blk.4.ffn_down_exps.weight q5_K -blk.4.ffn_down_shexp.weight q5_K +blk.4.ffn_down_exps.weight q5_1 +blk.4.ffn_down_shexp.weight q5_1 blk.4.attn_v.weight q5_K +blk.5.ffn_down_exps.weight iq4_nl +blk.5.ffn_down_shexp.weight iq4_nl blk.5.attn_v.weight q5_K +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_down_shexp.weight iq4_nl blk.6.attn_v.weight q5_K +blk.7.ffn_down_exps.weight iq4_nl +blk.7.ffn_down_shexp.weight iq4_nl blk.7.attn_v.weight q5_K +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_down_shexp.weight iq4_nl blk.8.attn_v.weight q5_K +blk.9.ffn_down_exps.weight iq4_nl +blk.9.ffn_down_shexp.weight iq4_nl blk.9.attn_v.weight q5_K +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_down_shexp.weight iq4_nl blk.10.attn_v.weight q5_K +blk.11.ffn_down_exps.weight iq4_nl +blk.11.ffn_down_shexp.weight iq4_nl blk.11.attn_v.weight q5_K +blk.12.ffn_down_exps.weight iq4_nl +blk.12.ffn_down_shexp.weight iq4_nl blk.12.attn_v.weight q5_K +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_down_shexp.weight iq4_nl blk.13.attn_v.weight q5_K +blk.14.ffn_down_exps.weight iq4_nl +blk.14.ffn_down_shexp.weight iq4_nl blk.14.attn_v.weight q5_K +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_down_shexp.weight iq4_nl blk.15.attn_v.weight q5_K +blk.16.ffn_down_exps.weight iq4_nl +blk.16.ffn_down_shexp.weight iq4_nl blk.16.attn_v.weight q5_K +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_down_shexp.weight iq4_nl blk.17.attn_v.weight q5_K +blk.18.ffn_down_exps.weight iq4_nl +blk.18.ffn_down_shexp.weight iq4_nl blk.18.attn_v.weight q5_K +blk.19.ffn_down_exps.weight iq4_nl +blk.19.ffn_down_shexp.weight iq4_nl blk.19.attn_v.weight q5_K +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_down_shexp.weight iq4_nl blk.20.attn_v.weight q5_K +blk.21.ffn_down_exps.weight iq4_nl +blk.21.ffn_down_shexp.weight iq4_nl blk.21.attn_v.weight q5_K +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_down_shexp.weight iq4_nl blk.22.attn_v.weight q5_K +blk.23.ffn_down_exps.weight iq4_nl +blk.23.ffn_down_shexp.weight iq4_nl blk.23.attn_v.weight q5_K +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_down_shexp.weight iq4_nl blk.24.attn_v.weight q5_K +blk.25.ffn_down_exps.weight iq4_nl +blk.25.ffn_down_shexp.weight iq4_nl blk.25.attn_v.weight q5_K +blk.26.ffn_down_exps.weight iq4_nl +blk.26.ffn_down_shexp.weight iq4_nl blk.26.attn_v.weight q5_K +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_down_shexp.weight iq4_nl blk.27.attn_v.weight q5_K +blk.28.ffn_down_exps.weight iq4_nl +blk.28.ffn_down_shexp.weight iq4_nl blk.28.attn_v.weight q5_K +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_down_shexp.weight iq4_nl blk.29.attn_v.weight q5_K +blk.30.ffn_down_exps.weight iq4_nl +blk.30.ffn_down_shexp.weight iq4_nl blk.30.attn_v.weight q5_K +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_down_shexp.weight iq4_nl blk.31.attn_v.weight q5_K +blk.32.ffn_down_exps.weight iq4_nl +blk.32.ffn_down_shexp.weight iq4_nl blk.32.attn_v.weight q5_K +blk.33.ffn_down_exps.weight iq4_nl +blk.33.ffn_down_shexp.weight iq4_nl blk.33.attn_v.weight q5_K +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_down_shexp.weight iq4_nl blk.34.attn_v.weight q5_K +blk.35.ffn_down_exps.weight iq4_nl +blk.35.ffn_down_shexp.weight iq4_nl blk.35.attn_v.weight q5_K +blk.36.ffn_down_exps.weight iq4_nl +blk.36.ffn_down_shexp.weight iq4_nl blk.36.attn_v.weight q5_K +blk.37.ffn_down_exps.weight iq4_nl +blk.37.ffn_down_shexp.weight iq4_nl blk.37.attn_v.weight q5_K +blk.38.ffn_down_exps.weight iq4_nl +blk.38.ffn_down_shexp.weight iq4_nl blk.38.attn_v.weight q5_K +blk.39.ffn_down_exps.weight iq4_nl +blk.39.ffn_down_shexp.weight iq4_nl blk.39.attn_v.weight q5_K +blk.40.ffn_down_exps.weight iq4_nl +blk.40.ffn_down_shexp.weight iq4_nl blk.40.attn_v.weight q5_K +blk.41.ffn_down_exps.weight iq4_nl +blk.41.ffn_down_shexp.weight iq4_nl blk.41.attn_v.weight q5_K +blk.42.ffn_down_exps.weight iq4_nl +blk.42.ffn_down_shexp.weight iq4_nl blk.42.attn_v.weight q5_K +blk.43.ffn_down_exps.weight iq4_nl +blk.43.ffn_down_shexp.weight iq4_nl blk.43.attn_v.weight q5_K +blk.44.ffn_down_exps.weight iq4_nl +blk.44.ffn_down_shexp.weight iq4_nl blk.44.attn_v.weight q5_K output.weight q6_K +blk.45.ffn_down_exps.weight iq4_nl +blk.45.ffn_down_shexp.weight iq4_nl blk.45.attn_v.weight q5_K [IQ1_M] iq1_m token_embd.weight q2_K -blk.0.ffn_down.weight q2_K +blk.0.ffn_down.weight q4_0 blk.0.attn_output.weight iq2_xxs blk.0.attn_v.weight q4_K -blk.1.ffn_down_exps.weight q2_K -blk.1.ffn_down_shexp.weight q2_K +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_down_shexp.weight q4_0 blk.1.attn_output.weight iq2_xxs blk.1.attn_v.weight q4_K -blk.2.ffn_down_exps.weight q2_K -blk.2.ffn_down_shexp.weight q2_K +blk.2.ffn_down_exps.weight q4_0 +blk.2.ffn_down_shexp.weight q4_0 blk.2.attn_output.weight iq2_xxs blk.2.attn_v.weight q4_K +blk.3.ffn_down_exps.weight iq4_nl +blk.3.ffn_down_shexp.weight iq4_nl blk.3.attn_output.weight iq2_xxs blk.3.attn_v.weight q4_K +blk.4.ffn_down_exps.weight iq4_nl +blk.4.ffn_down_shexp.weight iq4_nl blk.4.attn_output.weight iq2_xxs blk.4.attn_v.weight q4_K +blk.5.ffn_down_exps.weight iq4_nl +blk.5.ffn_down_shexp.weight iq4_nl blk.5.attn_output.weight iq2_xxs blk.5.attn_v.weight q4_K +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_down_shexp.weight iq4_nl blk.6.attn_output.weight iq2_xxs blk.6.attn_v.weight q4_K +blk.7.ffn_down_exps.weight iq4_nl +blk.7.ffn_down_shexp.weight iq4_nl blk.7.attn_output.weight iq2_xxs blk.7.attn_v.weight q4_K +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_down_shexp.weight iq4_nl blk.8.attn_output.weight iq2_xxs blk.8.attn_v.weight q4_K +blk.9.ffn_down_exps.weight iq4_nl +blk.9.ffn_down_shexp.weight iq4_nl blk.9.attn_output.weight iq2_xxs blk.9.attn_v.weight q4_K +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_down_shexp.weight iq4_nl blk.10.attn_output.weight iq2_xxs blk.10.attn_v.weight q4_K +blk.11.ffn_down_exps.weight iq4_nl +blk.11.ffn_down_shexp.weight iq4_nl blk.11.attn_output.weight iq2_xxs blk.11.attn_v.weight q4_K +blk.12.ffn_down_exps.weight iq4_nl +blk.12.ffn_down_shexp.weight iq4_nl blk.12.attn_output.weight iq2_xxs blk.12.attn_v.weight q4_K +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_down_shexp.weight iq4_nl blk.13.attn_output.weight iq2_xxs blk.13.attn_v.weight q4_K +blk.14.ffn_down_exps.weight iq4_nl +blk.14.ffn_down_shexp.weight iq4_nl blk.14.attn_output.weight iq2_xxs blk.14.attn_v.weight q4_K +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_down_shexp.weight iq4_nl blk.15.attn_output.weight iq2_xxs blk.15.attn_v.weight q4_K +blk.16.ffn_down_exps.weight iq4_nl +blk.16.ffn_down_shexp.weight iq4_nl blk.16.attn_output.weight iq2_xxs blk.16.attn_v.weight q4_K +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_down_shexp.weight iq4_nl blk.17.attn_output.weight iq2_xxs blk.17.attn_v.weight q4_K +blk.18.ffn_down_exps.weight iq4_nl +blk.18.ffn_down_shexp.weight iq4_nl blk.18.attn_output.weight iq2_xxs blk.18.attn_v.weight q4_K +blk.19.ffn_down_exps.weight iq4_nl +blk.19.ffn_down_shexp.weight iq4_nl blk.19.attn_output.weight iq2_xxs blk.19.attn_v.weight q4_K +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_down_shexp.weight iq4_nl blk.20.attn_output.weight iq2_xxs blk.20.attn_v.weight q4_K +blk.21.ffn_down_exps.weight iq4_nl +blk.21.ffn_down_shexp.weight iq4_nl blk.21.attn_output.weight iq2_xxs blk.21.attn_v.weight q4_K +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_down_shexp.weight iq4_nl blk.22.attn_output.weight iq2_xxs blk.22.attn_v.weight q4_K +blk.23.ffn_down_exps.weight iq4_nl +blk.23.ffn_down_shexp.weight iq4_nl blk.23.attn_output.weight iq2_xxs blk.23.attn_v.weight q4_K +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_down_shexp.weight iq4_nl blk.24.attn_output.weight iq2_xxs blk.24.attn_v.weight q4_K +blk.25.ffn_down_exps.weight iq4_nl +blk.25.ffn_down_shexp.weight iq4_nl blk.25.attn_output.weight iq2_xxs blk.25.attn_v.weight q4_K +blk.26.ffn_down_exps.weight iq4_nl +blk.26.ffn_down_shexp.weight iq4_nl blk.26.attn_output.weight iq2_xxs blk.26.attn_v.weight q4_K +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_down_shexp.weight iq4_nl blk.27.attn_output.weight iq2_xxs blk.27.attn_v.weight q4_K +blk.28.ffn_down_exps.weight iq4_nl +blk.28.ffn_down_shexp.weight iq4_nl blk.28.attn_output.weight iq2_xxs blk.28.attn_v.weight q4_K +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_down_shexp.weight iq4_nl blk.29.attn_output.weight iq2_xxs blk.29.attn_v.weight q4_K +blk.30.ffn_down_exps.weight iq4_nl +blk.30.ffn_down_shexp.weight iq4_nl blk.30.attn_output.weight iq2_xxs blk.30.attn_v.weight q4_K +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_down_shexp.weight iq4_nl blk.31.attn_output.weight iq2_xxs blk.31.attn_v.weight q4_K +blk.32.ffn_down_exps.weight iq4_nl +blk.32.ffn_down_shexp.weight iq4_nl blk.32.attn_output.weight iq2_xxs blk.32.attn_v.weight q4_K +blk.33.ffn_down_exps.weight iq4_nl +blk.33.ffn_down_shexp.weight iq4_nl blk.33.attn_output.weight iq2_xxs blk.33.attn_v.weight q4_K +blk.34.ffn_down_exps.weight iq4_nl blk.34.attn_output.weight iq2_xxs +blk.34.ffn_down_shexp.weight iq4_nl blk.34.attn_v.weight q4_K +blk.35.ffn_down_exps.weight iq4_nl +blk.35.ffn_down_shexp.weight iq4_nl blk.35.attn_output.weight iq2_xxs blk.35.attn_v.weight q4_K +blk.36.ffn_down_exps.weight iq4_nl +blk.36.ffn_down_shexp.weight iq4_nl blk.36.attn_output.weight iq2_xxs blk.36.attn_v.weight q4_K +blk.37.ffn_down_exps.weight iq4_nl +blk.37.ffn_down_shexp.weight iq4_nl blk.37.attn_output.weight iq2_xxs blk.37.attn_v.weight q4_K +blk.38.ffn_down_exps.weight iq4_nl +blk.38.ffn_down_shexp.weight iq4_nl blk.38.attn_output.weight iq2_xxs blk.38.attn_v.weight q4_K +blk.39.ffn_down_exps.weight iq4_nl +blk.39.ffn_down_shexp.weight iq4_nl blk.39.attn_output.weight iq2_xxs blk.39.attn_v.weight q4_K +blk.40.ffn_down_exps.weight iq4_nl +blk.40.ffn_down_shexp.weight iq4_nl blk.40.attn_output.weight iq2_xxs blk.40.attn_v.weight q4_K +blk.41.ffn_down_exps.weight iq4_nl +blk.41.ffn_down_shexp.weight iq4_nl blk.41.attn_output.weight iq2_xxs blk.41.attn_v.weight q4_K +blk.42.ffn_down_exps.weight iq4_nl +blk.42.ffn_down_shexp.weight iq4_nl blk.42.attn_v.weight q4_K blk.42.attn_output.weight iq2_xxs +blk.43.ffn_down_exps.weight iq4_nl +blk.43.ffn_down_shexp.weight iq4_nl blk.43.attn_output.weight iq2_xxs blk.43.attn_v.weight q4_K +blk.44.ffn_down_exps.weight iq4_nl +blk.44.ffn_down_shexp.weight iq4_nl blk.44.attn_output.weight iq2_xxs blk.44.attn_v.weight q4_K output.weight q5_K +blk.45.ffn_down_exps.weight iq4_nl +blk.45.ffn_down_shexp.weight iq4_nl blk.45.attn_output.weight iq2_xxs blk.45.attn_v.weight q4_K [BF16] bf16 -output.weight q6_K [TQ1_0] tq1_0 token_embd.weight q4_K +blk.0.ffn_down.weight q4_0 +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_down_shexp.weight q4_0 +blk.2.ffn_down_exps.weight q4_0 +blk.2.ffn_down_shexp.weight q4_0 +blk.3.ffn_down_exps.weight q4_0 +blk.3.ffn_down_shexp.weight q4_0 +blk.4.ffn_down_exps.weight q4_0 +blk.4.ffn_down_shexp.weight q4_0 +blk.5.ffn_down_exps.weight q4_0 +blk.5.ffn_down_shexp.weight q4_0 +blk.6.ffn_down_exps.weight q4_0 +blk.6.ffn_down_shexp.weight q4_0 +blk.7.ffn_down_exps.weight q4_0 +blk.7.ffn_down_shexp.weight q4_0 +blk.8.ffn_down_exps.weight q4_0 +blk.8.ffn_down_shexp.weight q4_0 +blk.9.ffn_down_exps.weight q4_0 +blk.9.ffn_down_shexp.weight q4_0 +blk.10.ffn_down_exps.weight q4_0 +blk.10.ffn_down_shexp.weight q4_0 +blk.11.ffn_down_exps.weight q4_0 +blk.11.ffn_down_shexp.weight q4_0 +blk.12.ffn_down_exps.weight q4_0 +blk.12.ffn_down_shexp.weight q4_0 +blk.13.ffn_down_exps.weight q4_0 +blk.13.ffn_down_shexp.weight q4_0 +blk.14.ffn_down_exps.weight q4_0 +blk.14.ffn_down_shexp.weight q4_0 +blk.15.ffn_down_exps.weight q4_0 +blk.15.ffn_down_shexp.weight q4_0 +blk.16.ffn_down_exps.weight q4_0 +blk.16.ffn_down_shexp.weight q4_0 +blk.17.ffn_down_exps.weight q4_0 +blk.17.ffn_down_shexp.weight q4_0 +blk.18.ffn_down_exps.weight q4_0 +blk.18.ffn_down_shexp.weight q4_0 +blk.19.ffn_down_exps.weight q4_0 +blk.19.ffn_down_shexp.weight q4_0 +blk.20.ffn_down_exps.weight q4_0 +blk.20.ffn_down_shexp.weight q4_0 +blk.21.ffn_down_exps.weight q4_0 +blk.21.ffn_down_shexp.weight q4_0 +blk.22.ffn_down_exps.weight q4_0 +blk.22.ffn_down_shexp.weight q4_0 +blk.23.ffn_down_exps.weight q4_0 +blk.23.ffn_down_shexp.weight q4_0 +blk.24.ffn_down_exps.weight q4_0 +blk.24.ffn_down_shexp.weight q4_0 +blk.25.ffn_down_exps.weight q4_0 +blk.25.ffn_down_shexp.weight q4_0 +blk.26.ffn_down_exps.weight q4_0 +blk.26.ffn_down_shexp.weight q4_0 +blk.27.ffn_down_exps.weight q4_0 +blk.27.ffn_down_shexp.weight q4_0 +blk.28.ffn_down_exps.weight q4_0 +blk.28.ffn_down_shexp.weight q4_0 +blk.29.ffn_down_exps.weight q4_0 +blk.29.ffn_down_shexp.weight q4_0 +blk.30.ffn_down_exps.weight q4_0 +blk.30.ffn_down_shexp.weight q4_0 +blk.31.ffn_down_exps.weight q4_0 +blk.31.ffn_down_shexp.weight q4_0 +blk.32.ffn_down_exps.weight q4_0 +blk.32.ffn_down_shexp.weight q4_0 +blk.33.ffn_down_exps.weight q4_0 +blk.33.ffn_down_shexp.weight q4_0 +blk.34.ffn_down_exps.weight q4_0 +blk.34.ffn_down_shexp.weight q4_0 +blk.35.ffn_down_exps.weight q4_0 +blk.35.ffn_down_shexp.weight q4_0 +blk.36.ffn_down_exps.weight q4_0 +blk.36.ffn_down_shexp.weight q4_0 +blk.37.ffn_down_exps.weight q4_0 +blk.37.ffn_down_shexp.weight q4_0 +blk.38.ffn_down_exps.weight q4_0 +blk.38.ffn_down_shexp.weight q4_0 +blk.39.ffn_down_exps.weight q4_0 +blk.39.ffn_down_shexp.weight q4_0 +blk.40.ffn_down_exps.weight q4_0 +blk.40.ffn_down_shexp.weight q4_0 +blk.41.ffn_down_exps.weight q4_0 +blk.41.ffn_down_shexp.weight q4_0 +blk.42.ffn_down_exps.weight q4_0 +blk.42.ffn_down_shexp.weight q4_0 +blk.43.ffn_down_exps.weight q4_0 +blk.43.ffn_down_shexp.weight q4_0 +blk.44.ffn_down_exps.weight q4_0 +blk.44.ffn_down_shexp.weight q4_0 output.weight q6_K +blk.45.ffn_down_exps.weight q4_0 +blk.45.ffn_down_shexp.weight q4_0 [TQ2_0] tq2_0 token_embd.weight q4_K +blk.0.ffn_down.weight q4_0 +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_down_shexp.weight q4_0 +blk.2.ffn_down_exps.weight q4_0 +blk.2.ffn_down_shexp.weight q4_0 +blk.3.ffn_down_exps.weight q4_0 +blk.3.ffn_down_shexp.weight q4_0 +blk.4.ffn_down_exps.weight q4_0 +blk.4.ffn_down_shexp.weight q4_0 +blk.5.ffn_down_exps.weight q4_0 +blk.5.ffn_down_shexp.weight q4_0 +blk.6.ffn_down_exps.weight q4_0 +blk.6.ffn_down_shexp.weight q4_0 +blk.7.ffn_down_exps.weight q4_0 +blk.7.ffn_down_shexp.weight q4_0 +blk.8.ffn_down_exps.weight q4_0 +blk.8.ffn_down_shexp.weight q4_0 +blk.9.ffn_down_exps.weight q4_0 +blk.9.ffn_down_shexp.weight q4_0 +blk.10.ffn_down_exps.weight q4_0 +blk.10.ffn_down_shexp.weight q4_0 +blk.11.ffn_down_exps.weight q4_0 +blk.11.ffn_down_shexp.weight q4_0 +blk.12.ffn_down_exps.weight q4_0 +blk.12.ffn_down_shexp.weight q4_0 +blk.13.ffn_down_exps.weight q4_0 +blk.13.ffn_down_shexp.weight q4_0 +blk.14.ffn_down_exps.weight q4_0 +blk.14.ffn_down_shexp.weight q4_0 +blk.15.ffn_down_exps.weight q4_0 +blk.15.ffn_down_shexp.weight q4_0 +blk.16.ffn_down_exps.weight q4_0 +blk.16.ffn_down_shexp.weight q4_0 +blk.17.ffn_down_exps.weight q4_0 +blk.17.ffn_down_shexp.weight q4_0 +blk.18.ffn_down_exps.weight q4_0 +blk.18.ffn_down_shexp.weight q4_0 +blk.19.ffn_down_exps.weight q4_0 +blk.19.ffn_down_shexp.weight q4_0 +blk.20.ffn_down_exps.weight q4_0 +blk.20.ffn_down_shexp.weight q4_0 +blk.21.ffn_down_exps.weight q4_0 +blk.21.ffn_down_shexp.weight q4_0 +blk.22.ffn_down_exps.weight q4_0 +blk.22.ffn_down_shexp.weight q4_0 +blk.23.ffn_down_exps.weight q4_0 +blk.23.ffn_down_shexp.weight q4_0 +blk.24.ffn_down_exps.weight q4_0 +blk.24.ffn_down_shexp.weight q4_0 +blk.25.ffn_down_exps.weight q4_0 +blk.25.ffn_down_shexp.weight q4_0 +blk.26.ffn_down_exps.weight q4_0 +blk.26.ffn_down_shexp.weight q4_0 +blk.27.ffn_down_exps.weight q4_0 +blk.27.ffn_down_shexp.weight q4_0 +blk.28.ffn_down_exps.weight q4_0 +blk.28.ffn_down_shexp.weight q4_0 +blk.29.ffn_down_exps.weight q4_0 +blk.29.ffn_down_shexp.weight q4_0 +blk.30.ffn_down_exps.weight q4_0 +blk.30.ffn_down_shexp.weight q4_0 +blk.31.ffn_down_exps.weight q4_0 +blk.31.ffn_down_shexp.weight q4_0 +blk.32.ffn_down_exps.weight q4_0 +blk.32.ffn_down_shexp.weight q4_0 +blk.33.ffn_down_exps.weight q4_0 +blk.33.ffn_down_shexp.weight q4_0 +blk.34.ffn_down_exps.weight q4_0 +blk.34.ffn_down_shexp.weight q4_0 +blk.35.ffn_down_exps.weight q4_0 +blk.35.ffn_down_shexp.weight q4_0 +blk.36.ffn_down_exps.weight q4_0 +blk.36.ffn_down_shexp.weight q4_0 +blk.37.ffn_down_exps.weight q4_0 +blk.37.ffn_down_shexp.weight q4_0 +blk.38.ffn_down_exps.weight q4_0 +blk.38.ffn_down_shexp.weight q4_0 +blk.39.ffn_down_exps.weight q4_0 +blk.39.ffn_down_shexp.weight q4_0 +blk.40.ffn_down_exps.weight q4_0 +blk.40.ffn_down_shexp.weight q4_0 +blk.41.ffn_down_exps.weight q4_0 +blk.41.ffn_down_shexp.weight q4_0 +blk.42.ffn_down_exps.weight q4_0 +blk.42.ffn_down_shexp.weight q4_0 +blk.43.ffn_down_exps.weight q4_0 +blk.43.ffn_down_shexp.weight q4_0 +blk.44.ffn_down_exps.weight q4_0 +blk.44.ffn_down_shexp.weight q4_0 output.weight q6_K +blk.45.ffn_down_exps.weight q4_0 +blk.45.ffn_down_shexp.weight q4_0 [MXFP4_MOE] mxfp4 token_embd.weight q8_0 diff --git a/tests/snapshots/gpt-oss-120b.schema b/tests/snapshots/gpt-oss-120b.schema index e66fbcfd2b..25c98f23cd 100644 --- a/tests/snapshots/gpt-oss-120b.schema +++ b/tests/snapshots/gpt-oss-120b.schema @@ -2,1470 +2,5451 @@ # n_embd=2880, n_ff=2880, n_vocab=0, n_layer=36, n_head=64, n_head_kv=8, n_expert=128 [F32] f32 -output.weight q6_K [F16] f16 -output.weight q6_K [Q4_0] q4_0 -output.weight q6_K +output.weight q8_0 [Q4_1] q4_1 -output.weight q6_K +output.weight q8_0 [Q8_0] q8_0 [Q5_0] q5_0 -output.weight q6_K +output.weight q8_0 [Q5_1] q5_1 -output.weight q6_K +output.weight q8_0 [Q2_K] q2_K output.weight q8_0 +token_embd.weight q4_0 +blk.0.attn_k.weight q4_0 blk.0.attn_output.weight q3_K -blk.0.attn_v.weight q4_K -blk.0.ffn_down_exps.weight q3_K +blk.0.attn_q.weight q4_0 +blk.0.attn_v.weight q5_0 +blk.0.ffn_down_exps.weight q4_0 +blk.0.ffn_gate_exps.weight q4_0 +blk.0.ffn_up_exps.weight q4_0 +blk.1.attn_k.weight q4_0 blk.1.attn_output.weight q3_K -blk.1.attn_v.weight q4_K -blk.1.ffn_down_exps.weight q3_K +blk.1.attn_q.weight q4_0 +blk.1.attn_v.weight q5_0 +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_gate_exps.weight q4_0 +blk.1.ffn_up_exps.weight q4_0 +blk.2.attn_k.weight q4_0 blk.2.attn_output.weight q3_K -blk.2.attn_v.weight q4_K -blk.2.ffn_down_exps.weight q3_K +blk.2.attn_q.weight q4_0 +blk.2.attn_v.weight q5_0 +blk.2.ffn_down_exps.weight q4_0 +blk.2.ffn_gate_exps.weight q4_0 +blk.2.ffn_up_exps.weight q4_0 +blk.3.attn_k.weight q4_0 blk.3.attn_output.weight q3_K -blk.3.attn_v.weight q4_K -blk.3.ffn_down_exps.weight q3_K +blk.3.attn_q.weight q4_0 +blk.3.attn_v.weight q5_0 +blk.3.ffn_down_exps.weight q4_0 +blk.3.ffn_gate_exps.weight q4_0 +blk.3.ffn_up_exps.weight q4_0 +blk.4.attn_k.weight q4_0 blk.4.attn_output.weight q3_K -blk.4.attn_v.weight q4_K -blk.4.ffn_down_exps.weight q3_K +blk.4.attn_q.weight q4_0 +blk.4.attn_v.weight q5_0 +blk.4.ffn_down_exps.weight q4_0 +blk.4.ffn_gate_exps.weight q4_0 +blk.4.ffn_up_exps.weight q4_0 +blk.5.attn_k.weight q4_0 blk.5.attn_output.weight q3_K -blk.5.attn_v.weight q4_K -blk.5.ffn_down_exps.weight q3_K +blk.5.attn_q.weight q4_0 +blk.5.attn_v.weight q5_0 +blk.5.ffn_down_exps.weight q4_0 +blk.5.ffn_gate_exps.weight q4_0 +blk.5.ffn_up_exps.weight q4_0 +blk.6.attn_k.weight q4_0 blk.6.attn_output.weight q3_K -blk.6.attn_v.weight q4_K -blk.6.ffn_down_exps.weight q3_K +blk.6.attn_q.weight q4_0 +blk.6.attn_v.weight q5_0 +blk.6.ffn_down_exps.weight q4_0 +blk.6.ffn_gate_exps.weight q4_0 +blk.6.ffn_up_exps.weight q4_0 +blk.7.attn_k.weight q4_0 blk.7.attn_output.weight q3_K -blk.7.attn_v.weight q4_K -blk.7.ffn_down_exps.weight q3_K +blk.7.attn_q.weight q4_0 +blk.7.attn_v.weight q5_0 +blk.7.ffn_down_exps.weight q4_0 +blk.7.ffn_gate_exps.weight q4_0 +blk.7.ffn_up_exps.weight q4_0 +blk.8.attn_k.weight q4_0 blk.8.attn_output.weight q3_K -blk.8.attn_v.weight q4_K -blk.8.ffn_down_exps.weight q3_K +blk.8.attn_q.weight q4_0 +blk.8.attn_v.weight q5_0 +blk.8.ffn_down_exps.weight q4_0 +blk.8.ffn_gate_exps.weight q4_0 +blk.8.ffn_up_exps.weight q4_0 +blk.9.attn_k.weight q4_0 blk.9.attn_output.weight q3_K -blk.9.attn_v.weight q4_K -blk.9.ffn_down_exps.weight q3_K +blk.9.attn_q.weight q4_0 +blk.9.attn_v.weight q5_0 +blk.9.ffn_down_exps.weight q4_0 +blk.9.ffn_gate_exps.weight q4_0 +blk.9.ffn_up_exps.weight q4_0 +blk.10.attn_k.weight q4_0 blk.10.attn_output.weight q3_K -blk.10.attn_v.weight q4_K -blk.10.ffn_down_exps.weight q3_K +blk.10.attn_q.weight q4_0 +blk.10.attn_v.weight q5_0 +blk.10.ffn_down_exps.weight q4_0 +blk.10.ffn_gate_exps.weight q4_0 +blk.10.ffn_up_exps.weight q4_0 +blk.11.attn_k.weight q4_0 blk.11.attn_output.weight q3_K -blk.11.attn_v.weight q4_K -blk.11.ffn_down_exps.weight q3_K +blk.11.attn_q.weight q4_0 +blk.11.attn_v.weight q5_0 +blk.11.ffn_down_exps.weight q4_0 +blk.11.ffn_gate_exps.weight q4_0 +blk.11.ffn_up_exps.weight q4_0 +blk.12.attn_k.weight q4_0 blk.12.attn_output.weight q3_K -blk.12.attn_v.weight q4_K -blk.12.ffn_down_exps.weight q3_K +blk.12.attn_q.weight q4_0 +blk.12.attn_v.weight q5_0 +blk.12.ffn_down_exps.weight q4_0 +blk.12.ffn_gate_exps.weight q4_0 +blk.12.ffn_up_exps.weight q4_0 +blk.13.attn_k.weight q4_0 blk.13.attn_output.weight q3_K -blk.13.attn_v.weight q4_K -blk.13.ffn_down_exps.weight q3_K +blk.13.attn_q.weight q4_0 +blk.13.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight q4_0 +blk.13.ffn_gate_exps.weight q4_0 +blk.13.ffn_up_exps.weight q4_0 +blk.14.attn_k.weight q4_0 blk.14.attn_output.weight q3_K -blk.14.attn_v.weight q4_K -blk.14.ffn_down_exps.weight q3_K +blk.14.attn_q.weight q4_0 +blk.14.attn_v.weight q5_0 +blk.14.ffn_down_exps.weight q4_0 +blk.14.ffn_gate_exps.weight q4_0 +blk.14.ffn_up_exps.weight q4_0 +blk.15.attn_k.weight q4_0 blk.15.attn_output.weight q3_K -blk.15.attn_v.weight q4_K -blk.15.ffn_down_exps.weight q3_K +blk.15.attn_q.weight q4_0 +blk.15.attn_v.weight q5_0 +blk.15.ffn_down_exps.weight q4_0 +blk.15.ffn_gate_exps.weight q4_0 +blk.15.ffn_up_exps.weight q4_0 +blk.16.attn_k.weight q4_0 blk.16.attn_output.weight q3_K -blk.16.attn_v.weight q4_K -blk.16.ffn_down_exps.weight q3_K +blk.16.attn_q.weight q4_0 +blk.16.attn_v.weight q5_0 +blk.16.ffn_down_exps.weight q4_0 +blk.16.ffn_gate_exps.weight q4_0 +blk.16.ffn_up_exps.weight q4_0 +blk.17.attn_k.weight q4_0 blk.17.attn_output.weight q3_K -blk.17.attn_v.weight q4_K -blk.17.ffn_down_exps.weight q3_K +blk.17.attn_q.weight q4_0 +blk.17.attn_v.weight q5_0 +blk.17.ffn_down_exps.weight q4_0 +blk.17.ffn_gate_exps.weight q4_0 +blk.17.ffn_up_exps.weight q4_0 +blk.18.attn_k.weight q4_0 blk.18.attn_output.weight q3_K -blk.18.attn_v.weight q4_K -blk.18.ffn_down_exps.weight q3_K +blk.18.attn_q.weight q4_0 +blk.18.attn_v.weight q5_0 +blk.18.ffn_down_exps.weight q4_0 +blk.18.ffn_gate_exps.weight q4_0 +blk.18.ffn_up_exps.weight q4_0 +blk.19.attn_k.weight q4_0 blk.19.attn_output.weight q3_K -blk.19.attn_v.weight q4_K -blk.19.ffn_down_exps.weight q3_K +blk.19.attn_q.weight q4_0 +blk.19.attn_v.weight q5_0 +blk.19.ffn_down_exps.weight q4_0 +blk.19.ffn_gate_exps.weight q4_0 +blk.19.ffn_up_exps.weight q4_0 +blk.20.attn_k.weight q4_0 blk.20.attn_output.weight q3_K -blk.20.attn_v.weight q4_K -blk.20.ffn_down_exps.weight q3_K +blk.20.attn_q.weight q4_0 +blk.20.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight q4_0 +blk.20.ffn_gate_exps.weight q4_0 +blk.20.ffn_up_exps.weight q4_0 +blk.21.attn_k.weight q4_0 blk.21.attn_output.weight q3_K -blk.21.attn_v.weight q4_K -blk.21.ffn_down_exps.weight q3_K +blk.21.attn_q.weight q4_0 +blk.21.attn_v.weight q5_0 +blk.21.ffn_down_exps.weight q4_0 +blk.21.ffn_gate_exps.weight q4_0 +blk.21.ffn_up_exps.weight q4_0 +blk.22.attn_k.weight q4_0 blk.22.attn_output.weight q3_K -blk.22.attn_v.weight q4_K -blk.22.ffn_down_exps.weight q3_K +blk.22.attn_q.weight q4_0 +blk.22.attn_v.weight q5_0 +blk.22.ffn_down_exps.weight q4_0 +blk.22.ffn_gate_exps.weight q4_0 +blk.22.ffn_up_exps.weight q4_0 +blk.23.attn_k.weight q4_0 blk.23.attn_output.weight q3_K -blk.23.attn_v.weight q4_K -blk.23.ffn_down_exps.weight q3_K +blk.23.attn_q.weight q4_0 +blk.23.attn_v.weight q5_0 +blk.23.ffn_down_exps.weight q4_0 +blk.23.ffn_gate_exps.weight q4_0 +blk.23.ffn_up_exps.weight q4_0 +blk.24.attn_k.weight q4_0 blk.24.attn_output.weight q3_K -blk.24.attn_v.weight q4_K -blk.24.ffn_down_exps.weight q3_K +blk.24.attn_q.weight q4_0 +blk.24.attn_v.weight q5_0 +blk.24.ffn_down_exps.weight q4_0 +blk.24.ffn_gate_exps.weight q4_0 +blk.24.ffn_up_exps.weight q4_0 +blk.25.attn_k.weight q4_0 blk.25.attn_output.weight q3_K -blk.25.attn_v.weight q4_K -blk.25.ffn_down_exps.weight q3_K +blk.25.attn_q.weight q4_0 +blk.25.attn_v.weight q5_0 +blk.25.ffn_down_exps.weight q4_0 +blk.25.ffn_gate_exps.weight q4_0 +blk.25.ffn_up_exps.weight q4_0 +blk.26.attn_k.weight q4_0 blk.26.attn_output.weight q3_K -blk.26.attn_v.weight q4_K -blk.26.ffn_down_exps.weight q3_K +blk.26.attn_q.weight q4_0 +blk.26.attn_v.weight q5_0 +blk.26.ffn_down_exps.weight q4_0 +blk.26.ffn_gate_exps.weight q4_0 +blk.26.ffn_up_exps.weight q4_0 +blk.27.attn_k.weight q4_0 blk.27.attn_output.weight q3_K -blk.27.attn_v.weight q4_K -blk.27.ffn_down_exps.weight q3_K +blk.27.attn_q.weight q4_0 +blk.27.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight q4_0 +blk.27.ffn_gate_exps.weight q4_0 +blk.27.ffn_up_exps.weight q4_0 +blk.28.attn_k.weight q4_0 blk.28.attn_output.weight q3_K -blk.28.attn_v.weight q4_K -blk.28.ffn_down_exps.weight q3_K +blk.28.attn_q.weight q4_0 +blk.28.attn_v.weight q5_0 +blk.28.ffn_down_exps.weight q4_0 +blk.28.ffn_gate_exps.weight q4_0 +blk.28.ffn_up_exps.weight q4_0 +blk.29.attn_k.weight q4_0 blk.29.attn_output.weight q3_K -blk.29.attn_v.weight q4_K -blk.29.ffn_down_exps.weight q3_K +blk.29.attn_q.weight q4_0 +blk.29.attn_v.weight q5_0 +blk.29.ffn_down_exps.weight q4_0 +blk.29.ffn_gate_exps.weight q4_0 +blk.29.ffn_up_exps.weight q4_0 +blk.30.attn_k.weight q4_0 blk.30.attn_output.weight q3_K -blk.30.attn_v.weight q4_K -blk.30.ffn_down_exps.weight q3_K +blk.30.attn_q.weight q4_0 +blk.30.attn_v.weight q5_0 +blk.30.ffn_down_exps.weight q4_0 +blk.30.ffn_gate_exps.weight q4_0 +blk.30.ffn_up_exps.weight q4_0 +blk.31.attn_k.weight q4_0 blk.31.attn_output.weight q3_K -blk.31.attn_v.weight q4_K -blk.31.ffn_down_exps.weight q3_K +blk.31.attn_q.weight q4_0 +blk.31.attn_v.weight q5_0 +blk.31.ffn_down_exps.weight q4_0 +blk.31.ffn_gate_exps.weight q4_0 +blk.31.ffn_up_exps.weight q4_0 +blk.32.attn_k.weight q4_0 blk.32.attn_output.weight q3_K -blk.32.attn_v.weight q4_K -blk.32.ffn_down_exps.weight q3_K +blk.32.attn_q.weight q4_0 +blk.32.attn_v.weight q5_0 +blk.32.ffn_down_exps.weight q4_0 +blk.32.ffn_gate_exps.weight q4_0 +blk.32.ffn_up_exps.weight q4_0 +blk.33.attn_k.weight q4_0 blk.33.attn_output.weight q3_K -blk.33.attn_v.weight q4_K -blk.33.ffn_down_exps.weight q3_K +blk.33.attn_q.weight q4_0 +blk.33.attn_v.weight q5_0 +blk.33.ffn_down_exps.weight q4_0 +blk.33.ffn_gate_exps.weight q4_0 +blk.33.ffn_up_exps.weight q4_0 +blk.34.attn_k.weight q4_0 blk.34.attn_output.weight q3_K -blk.34.attn_v.weight q4_K -blk.34.ffn_down_exps.weight q3_K +blk.34.attn_q.weight q4_0 +blk.34.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight q4_0 +blk.34.ffn_gate_exps.weight q4_0 +blk.34.ffn_up_exps.weight q4_0 +blk.35.attn_k.weight q4_0 blk.35.attn_output.weight q3_K -blk.35.attn_v.weight q4_K -blk.35.ffn_down_exps.weight q3_K +blk.35.attn_q.weight q4_0 +blk.35.attn_v.weight q5_0 +blk.35.ffn_down_exps.weight q4_0 +blk.35.ffn_gate_exps.weight q4_0 +blk.35.ffn_up_exps.weight q4_0 [Q3_K_S] q3_K output.weight q8_0 +token_embd.weight q4_0 +blk.0.attn_k.weight q4_0 +blk.0.attn_q.weight q4_0 +blk.0.attn_v.weight q4_0 +blk.0.ffn_down_exps.weight q4_0 +blk.0.ffn_gate_exps.weight q4_0 +blk.0.ffn_up_exps.weight q4_0 +blk.1.attn_k.weight q4_0 +blk.1.attn_q.weight q4_0 +blk.1.attn_v.weight q4_0 +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_gate_exps.weight q4_0 +blk.1.ffn_up_exps.weight q4_0 +blk.2.attn_k.weight q4_0 +blk.2.attn_q.weight q4_0 +blk.2.attn_v.weight q4_0 +blk.2.ffn_down_exps.weight q4_0 +blk.2.ffn_gate_exps.weight q4_0 +blk.2.ffn_up_exps.weight q4_0 +blk.3.attn_k.weight q4_0 +blk.3.attn_q.weight q4_0 +blk.3.attn_v.weight q4_0 +blk.3.ffn_down_exps.weight q4_0 +blk.3.ffn_gate_exps.weight q4_0 +blk.3.ffn_up_exps.weight q4_0 +blk.4.attn_k.weight q4_0 +blk.4.attn_q.weight q4_0 +blk.4.attn_v.weight q4_0 +blk.4.ffn_down_exps.weight q4_0 +blk.4.ffn_gate_exps.weight q4_0 +blk.4.ffn_up_exps.weight q4_0 +blk.5.attn_k.weight q4_0 +blk.5.attn_q.weight q4_0 +blk.5.attn_v.weight q4_0 +blk.5.ffn_down_exps.weight q4_0 +blk.5.ffn_gate_exps.weight q4_0 +blk.5.ffn_up_exps.weight q4_0 +blk.6.attn_k.weight q4_0 +blk.6.attn_q.weight q4_0 +blk.6.attn_v.weight q4_0 +blk.6.ffn_down_exps.weight q4_0 +blk.6.ffn_gate_exps.weight q4_0 +blk.6.ffn_up_exps.weight q4_0 +blk.7.attn_k.weight q4_0 +blk.7.attn_q.weight q4_0 +blk.7.attn_v.weight q4_0 +blk.7.ffn_down_exps.weight q4_0 +blk.7.ffn_gate_exps.weight q4_0 +blk.7.ffn_up_exps.weight q4_0 +blk.8.attn_k.weight q4_0 +blk.8.attn_q.weight q4_0 +blk.8.attn_v.weight q4_0 +blk.8.ffn_down_exps.weight q4_0 +blk.8.ffn_gate_exps.weight q4_0 +blk.8.ffn_up_exps.weight q4_0 +blk.9.attn_k.weight q4_0 +blk.9.attn_q.weight q4_0 +blk.9.attn_v.weight q4_0 +blk.9.ffn_down_exps.weight q4_0 +blk.9.ffn_gate_exps.weight q4_0 +blk.9.ffn_up_exps.weight q4_0 +blk.10.attn_k.weight q4_0 +blk.10.attn_q.weight q4_0 +blk.10.attn_v.weight q4_0 +blk.10.ffn_down_exps.weight q4_0 +blk.10.ffn_gate_exps.weight q4_0 +blk.10.ffn_up_exps.weight q4_0 +blk.11.attn_k.weight q4_0 +blk.11.attn_q.weight q4_0 +blk.11.attn_v.weight q4_0 +blk.11.ffn_down_exps.weight q4_0 +blk.11.ffn_gate_exps.weight q4_0 +blk.11.ffn_up_exps.weight q4_0 +blk.12.attn_k.weight q4_0 +blk.12.attn_q.weight q4_0 +blk.12.attn_v.weight q4_0 +blk.12.ffn_down_exps.weight q4_0 +blk.12.ffn_gate_exps.weight q4_0 +blk.12.ffn_up_exps.weight q4_0 +blk.13.attn_k.weight q4_0 +blk.13.attn_q.weight q4_0 +blk.13.attn_v.weight q4_0 +blk.13.ffn_down_exps.weight q4_0 +blk.13.ffn_gate_exps.weight q4_0 +blk.13.ffn_up_exps.weight q4_0 +blk.14.attn_k.weight q4_0 +blk.14.attn_q.weight q4_0 +blk.14.attn_v.weight q4_0 +blk.14.ffn_down_exps.weight q4_0 +blk.14.ffn_gate_exps.weight q4_0 +blk.14.ffn_up_exps.weight q4_0 +blk.15.attn_k.weight q4_0 +blk.15.attn_q.weight q4_0 +blk.15.attn_v.weight q4_0 +blk.15.ffn_down_exps.weight q4_0 +blk.15.ffn_gate_exps.weight q4_0 +blk.15.ffn_up_exps.weight q4_0 +blk.16.attn_k.weight q4_0 +blk.16.attn_q.weight q4_0 +blk.16.attn_v.weight q4_0 +blk.16.ffn_down_exps.weight q4_0 +blk.16.ffn_gate_exps.weight q4_0 +blk.16.ffn_up_exps.weight q4_0 +blk.17.attn_k.weight q4_0 +blk.17.attn_q.weight q4_0 +blk.17.attn_v.weight q4_0 +blk.17.ffn_down_exps.weight q4_0 +blk.17.ffn_gate_exps.weight q4_0 +blk.17.ffn_up_exps.weight q4_0 +blk.18.attn_k.weight q4_0 +blk.18.attn_q.weight q4_0 +blk.18.attn_v.weight q4_0 +blk.18.ffn_down_exps.weight q4_0 +blk.18.ffn_gate_exps.weight q4_0 +blk.18.ffn_up_exps.weight q4_0 +blk.19.attn_k.weight q4_0 +blk.19.attn_q.weight q4_0 +blk.19.attn_v.weight q4_0 +blk.19.ffn_down_exps.weight q4_0 +blk.19.ffn_gate_exps.weight q4_0 +blk.19.ffn_up_exps.weight q4_0 +blk.20.attn_k.weight q4_0 +blk.20.attn_q.weight q4_0 +blk.20.attn_v.weight q4_0 +blk.20.ffn_down_exps.weight q4_0 +blk.20.ffn_gate_exps.weight q4_0 +blk.20.ffn_up_exps.weight q4_0 +blk.21.attn_k.weight q4_0 +blk.21.attn_q.weight q4_0 +blk.21.attn_v.weight q4_0 +blk.21.ffn_down_exps.weight q4_0 +blk.21.ffn_gate_exps.weight q4_0 +blk.21.ffn_up_exps.weight q4_0 +blk.22.attn_k.weight q4_0 +blk.22.attn_q.weight q4_0 +blk.22.attn_v.weight q4_0 +blk.22.ffn_down_exps.weight q4_0 +blk.22.ffn_gate_exps.weight q4_0 +blk.22.ffn_up_exps.weight q4_0 +blk.23.attn_k.weight q4_0 +blk.23.attn_q.weight q4_0 +blk.23.attn_v.weight q4_0 +blk.23.ffn_down_exps.weight q4_0 +blk.23.ffn_gate_exps.weight q4_0 +blk.23.ffn_up_exps.weight q4_0 +blk.24.attn_k.weight q4_0 +blk.24.attn_q.weight q4_0 +blk.24.attn_v.weight q4_0 +blk.24.ffn_down_exps.weight q4_0 +blk.24.ffn_gate_exps.weight q4_0 +blk.24.ffn_up_exps.weight q4_0 +blk.25.attn_k.weight q4_0 +blk.25.attn_q.weight q4_0 +blk.25.attn_v.weight q4_0 +blk.25.ffn_down_exps.weight q4_0 +blk.25.ffn_gate_exps.weight q4_0 +blk.25.ffn_up_exps.weight q4_0 +blk.26.attn_k.weight q4_0 +blk.26.attn_q.weight q4_0 +blk.26.attn_v.weight q4_0 +blk.26.ffn_down_exps.weight q4_0 +blk.26.ffn_gate_exps.weight q4_0 +blk.26.ffn_up_exps.weight q4_0 +blk.27.attn_k.weight q4_0 +blk.27.attn_q.weight q4_0 +blk.27.attn_v.weight q4_0 +blk.27.ffn_down_exps.weight q4_0 +blk.27.ffn_gate_exps.weight q4_0 +blk.27.ffn_up_exps.weight q4_0 +blk.28.attn_k.weight q4_0 +blk.28.attn_q.weight q4_0 +blk.28.attn_v.weight q4_0 +blk.28.ffn_down_exps.weight q4_0 +blk.28.ffn_gate_exps.weight q4_0 +blk.28.ffn_up_exps.weight q4_0 +blk.29.attn_k.weight q4_0 +blk.29.attn_q.weight q4_0 +blk.29.attn_v.weight q4_0 +blk.29.ffn_down_exps.weight q4_0 +blk.29.ffn_gate_exps.weight q4_0 +blk.29.ffn_up_exps.weight q4_0 +blk.30.attn_k.weight q4_0 +blk.30.attn_q.weight q4_0 +blk.30.attn_v.weight q4_0 +blk.30.ffn_down_exps.weight q4_0 +blk.30.ffn_gate_exps.weight q4_0 +blk.30.ffn_up_exps.weight q4_0 +blk.31.attn_k.weight q4_0 +blk.31.attn_q.weight q4_0 +blk.31.attn_v.weight q4_0 +blk.31.ffn_down_exps.weight q4_0 +blk.31.ffn_gate_exps.weight q4_0 +blk.31.ffn_up_exps.weight q4_0 +blk.32.attn_k.weight q4_0 +blk.32.attn_q.weight q4_0 +blk.32.attn_v.weight q4_0 +blk.32.ffn_down_exps.weight q4_0 +blk.32.ffn_gate_exps.weight q4_0 +blk.32.ffn_up_exps.weight q4_0 +blk.33.attn_k.weight q4_0 +blk.33.attn_q.weight q4_0 +blk.33.attn_v.weight q4_0 +blk.33.ffn_down_exps.weight q4_0 +blk.33.ffn_gate_exps.weight q4_0 +blk.33.ffn_up_exps.weight q4_0 +blk.34.attn_k.weight q4_0 +blk.34.attn_q.weight q4_0 +blk.34.attn_v.weight q4_0 +blk.34.ffn_down_exps.weight q4_0 +blk.34.ffn_gate_exps.weight q4_0 +blk.34.ffn_up_exps.weight q4_0 +blk.35.attn_k.weight q4_0 +blk.35.attn_q.weight q4_0 +blk.35.attn_v.weight q4_0 +blk.35.ffn_down_exps.weight q4_0 +blk.35.ffn_gate_exps.weight q4_0 +blk.35.ffn_up_exps.weight q4_0 [Q3_K_M] q3_K output.weight q8_0 +token_embd.weight q4_0 +blk.0.attn_k.weight q4_0 blk.0.attn_output.weight q4_K -blk.0.attn_v.weight q5_K -blk.0.ffn_down_exps.weight q5_K +blk.0.attn_q.weight q4_0 +blk.0.attn_v.weight q5_1 +blk.0.ffn_down_exps.weight q5_1 +blk.0.ffn_gate_exps.weight q4_0 +blk.0.ffn_up_exps.weight q4_0 +blk.1.attn_k.weight q4_0 blk.1.attn_output.weight q4_K -blk.1.attn_v.weight q5_K -blk.1.ffn_down_exps.weight q5_K +blk.1.attn_q.weight q4_0 +blk.1.attn_v.weight q5_1 +blk.1.ffn_down_exps.weight q5_1 +blk.1.ffn_gate_exps.weight q4_0 +blk.1.ffn_up_exps.weight q4_0 +blk.2.attn_k.weight q4_0 blk.2.attn_output.weight q4_K -blk.2.attn_v.weight q4_K -blk.2.ffn_down_exps.weight q4_K +blk.2.attn_q.weight q4_0 +blk.2.attn_v.weight q5_0 +blk.2.ffn_down_exps.weight q5_0 +blk.2.ffn_gate_exps.weight q4_0 +blk.2.ffn_up_exps.weight q4_0 +blk.3.attn_k.weight q4_0 blk.3.attn_output.weight q4_K -blk.3.attn_v.weight q4_K -blk.3.ffn_down_exps.weight q4_K +blk.3.attn_q.weight q4_0 +blk.3.attn_v.weight q5_0 +blk.3.ffn_down_exps.weight q5_0 +blk.3.ffn_gate_exps.weight q4_0 +blk.3.ffn_up_exps.weight q4_0 +blk.4.attn_k.weight q4_0 blk.4.attn_output.weight q4_K -blk.4.attn_v.weight q4_K -blk.4.ffn_down_exps.weight q4_K +blk.4.attn_q.weight q4_0 +blk.4.attn_v.weight q5_0 +blk.4.ffn_down_exps.weight q5_0 +blk.4.ffn_gate_exps.weight q4_0 +blk.4.ffn_up_exps.weight q4_0 +blk.5.attn_k.weight q4_0 blk.5.attn_output.weight q4_K -blk.5.attn_v.weight q4_K -blk.5.ffn_down_exps.weight q4_K +blk.5.attn_q.weight q4_0 +blk.5.attn_v.weight q5_0 +blk.5.ffn_down_exps.weight q5_0 +blk.5.ffn_gate_exps.weight q4_0 +blk.5.ffn_up_exps.weight q4_0 +blk.6.attn_k.weight q4_0 blk.6.attn_output.weight q4_K -blk.6.attn_v.weight q4_K -blk.6.ffn_down_exps.weight q4_K +blk.6.attn_q.weight q4_0 +blk.6.attn_v.weight q5_0 +blk.6.ffn_down_exps.weight q5_0 +blk.6.ffn_gate_exps.weight q4_0 +blk.6.ffn_up_exps.weight q4_0 +blk.7.attn_k.weight q4_0 blk.7.attn_output.weight q4_K -blk.7.attn_v.weight q4_K -blk.7.ffn_down_exps.weight q4_K +blk.7.attn_q.weight q4_0 +blk.7.attn_v.weight q5_0 +blk.7.ffn_down_exps.weight q5_0 +blk.7.ffn_gate_exps.weight q4_0 +blk.7.ffn_up_exps.weight q4_0 +blk.8.attn_k.weight q4_0 blk.8.attn_output.weight q4_K -blk.8.attn_v.weight q4_K -blk.8.ffn_down_exps.weight q4_K +blk.8.attn_q.weight q4_0 +blk.8.attn_v.weight q5_0 +blk.8.ffn_down_exps.weight q5_0 +blk.8.ffn_gate_exps.weight q4_0 +blk.8.ffn_up_exps.weight q4_0 +blk.9.attn_k.weight q4_0 blk.9.attn_output.weight q4_K -blk.9.attn_v.weight q4_K -blk.9.ffn_down_exps.weight q4_K +blk.9.attn_q.weight q4_0 +blk.9.attn_v.weight q5_0 +blk.9.ffn_down_exps.weight q5_0 +blk.9.ffn_gate_exps.weight q4_0 +blk.9.ffn_up_exps.weight q4_0 +blk.10.attn_k.weight q4_0 blk.10.attn_output.weight q4_K -blk.10.attn_v.weight q4_K -blk.10.ffn_down_exps.weight q4_K +blk.10.attn_q.weight q4_0 +blk.10.attn_v.weight q5_0 +blk.10.ffn_down_exps.weight q5_0 +blk.10.ffn_gate_exps.weight q4_0 +blk.10.ffn_up_exps.weight q4_0 +blk.11.attn_k.weight q4_0 blk.11.attn_output.weight q4_K -blk.11.attn_v.weight q4_K -blk.11.ffn_down_exps.weight q4_K +blk.11.attn_q.weight q4_0 +blk.11.attn_v.weight q5_0 +blk.11.ffn_down_exps.weight q5_0 +blk.11.ffn_gate_exps.weight q4_0 +blk.11.ffn_up_exps.weight q4_0 +blk.12.attn_k.weight q4_0 blk.12.attn_output.weight q4_K -blk.12.attn_v.weight q4_K -blk.12.ffn_down_exps.weight q4_K +blk.12.attn_q.weight q4_0 +blk.12.attn_v.weight q5_0 +blk.12.ffn_down_exps.weight q5_0 +blk.12.ffn_gate_exps.weight q4_0 +blk.12.ffn_up_exps.weight q4_0 +blk.13.attn_k.weight q4_0 blk.13.attn_output.weight q4_K -blk.13.attn_v.weight q4_K -blk.13.ffn_down_exps.weight q4_K +blk.13.attn_q.weight q4_0 +blk.13.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight q5_0 +blk.13.ffn_gate_exps.weight q4_0 +blk.13.ffn_up_exps.weight q4_0 +blk.14.attn_k.weight q4_0 blk.14.attn_output.weight q4_K -blk.14.attn_v.weight q4_K -blk.14.ffn_down_exps.weight q4_K +blk.14.attn_q.weight q4_0 +blk.14.attn_v.weight q5_0 +blk.14.ffn_down_exps.weight q5_0 +blk.14.ffn_gate_exps.weight q4_0 +blk.14.ffn_up_exps.weight q4_0 +blk.15.attn_k.weight q4_0 blk.15.attn_output.weight q4_K -blk.15.attn_v.weight q4_K -blk.15.ffn_down_exps.weight q4_K +blk.15.attn_q.weight q4_0 +blk.15.attn_v.weight q5_0 +blk.15.ffn_down_exps.weight q5_0 +blk.15.ffn_gate_exps.weight q4_0 +blk.15.ffn_up_exps.weight q4_0 +blk.16.attn_k.weight q4_0 blk.16.attn_output.weight q4_K -blk.16.attn_v.weight q4_K -blk.16.ffn_down_exps.weight q4_K +blk.16.attn_q.weight q4_0 +blk.16.attn_v.weight q5_0 +blk.16.ffn_down_exps.weight q5_0 +blk.16.ffn_gate_exps.weight q4_0 +blk.16.ffn_up_exps.weight q4_0 +blk.17.attn_k.weight q4_0 blk.17.attn_output.weight q4_K -blk.17.attn_v.weight q4_K -blk.17.ffn_down_exps.weight q4_K +blk.17.attn_q.weight q4_0 +blk.17.attn_v.weight q5_0 +blk.17.ffn_down_exps.weight q5_0 +blk.17.ffn_gate_exps.weight q4_0 +blk.17.ffn_up_exps.weight q4_0 +blk.18.attn_k.weight q4_0 blk.18.attn_output.weight q4_K -blk.18.attn_v.weight q4_K -blk.18.ffn_down_exps.weight q4_K +blk.18.attn_q.weight q4_0 +blk.18.attn_v.weight q5_0 +blk.18.ffn_down_exps.weight q5_0 +blk.18.ffn_gate_exps.weight q4_0 +blk.18.ffn_up_exps.weight q4_0 +blk.19.attn_k.weight q4_0 blk.19.attn_output.weight q4_K -blk.19.attn_v.weight q4_K -blk.19.ffn_down_exps.weight q4_K +blk.19.attn_q.weight q4_0 +blk.19.attn_v.weight q5_0 +blk.19.ffn_down_exps.weight q5_0 +blk.19.ffn_gate_exps.weight q4_0 +blk.19.ffn_up_exps.weight q4_0 +blk.20.attn_k.weight q4_0 blk.20.attn_output.weight q4_K -blk.20.attn_v.weight q4_K -blk.20.ffn_down_exps.weight q4_K +blk.20.attn_q.weight q4_0 +blk.20.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight q5_0 +blk.20.ffn_gate_exps.weight q4_0 +blk.20.ffn_up_exps.weight q4_0 +blk.21.attn_k.weight q4_0 blk.21.attn_output.weight q4_K -blk.21.attn_v.weight q4_K -blk.21.ffn_down_exps.weight q4_K +blk.21.attn_q.weight q4_0 +blk.21.attn_v.weight q5_0 +blk.21.ffn_down_exps.weight q5_0 +blk.21.ffn_gate_exps.weight q4_0 +blk.21.ffn_up_exps.weight q4_0 +blk.22.attn_k.weight q4_0 blk.22.attn_output.weight q4_K -blk.22.attn_v.weight q4_K -blk.22.ffn_down_exps.weight q4_K +blk.22.attn_q.weight q4_0 +blk.22.attn_v.weight q5_0 +blk.22.ffn_down_exps.weight q5_0 +blk.22.ffn_gate_exps.weight q4_0 +blk.22.ffn_up_exps.weight q4_0 +blk.23.attn_k.weight q4_0 blk.23.attn_output.weight q4_K -blk.23.attn_v.weight q4_K -blk.23.ffn_down_exps.weight q4_K +blk.23.attn_q.weight q4_0 +blk.23.attn_v.weight q5_0 +blk.23.ffn_down_exps.weight q5_0 +blk.23.ffn_gate_exps.weight q4_0 +blk.23.ffn_up_exps.weight q4_0 +blk.24.attn_k.weight q4_0 blk.24.attn_output.weight q4_K -blk.24.attn_v.weight q4_K -blk.24.ffn_down_exps.weight q4_K +blk.24.attn_q.weight q4_0 +blk.24.attn_v.weight q5_0 +blk.24.ffn_down_exps.weight q5_0 +blk.24.ffn_gate_exps.weight q4_0 +blk.24.ffn_up_exps.weight q4_0 +blk.25.attn_k.weight q4_0 blk.25.attn_output.weight q4_K -blk.25.attn_v.weight q4_K -blk.25.ffn_down_exps.weight q4_K +blk.25.attn_q.weight q4_0 +blk.25.attn_v.weight q5_0 +blk.25.ffn_down_exps.weight q5_0 +blk.25.ffn_gate_exps.weight q4_0 +blk.25.ffn_up_exps.weight q4_0 +blk.26.attn_k.weight q4_0 blk.26.attn_output.weight q4_K -blk.26.attn_v.weight q4_K -blk.26.ffn_down_exps.weight q4_K +blk.26.attn_q.weight q4_0 +blk.26.attn_v.weight q5_0 +blk.26.ffn_down_exps.weight q5_0 +blk.26.ffn_gate_exps.weight q4_0 +blk.26.ffn_up_exps.weight q4_0 +blk.27.attn_k.weight q4_0 blk.27.attn_output.weight q4_K -blk.27.attn_v.weight q4_K -blk.27.ffn_down_exps.weight q4_K +blk.27.attn_q.weight q4_0 +blk.27.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight q5_0 +blk.27.ffn_gate_exps.weight q4_0 +blk.27.ffn_up_exps.weight q4_0 +blk.28.attn_k.weight q4_0 blk.28.attn_output.weight q4_K -blk.28.attn_v.weight q4_K -blk.28.ffn_down_exps.weight q4_K +blk.28.attn_q.weight q4_0 +blk.28.attn_v.weight q5_0 +blk.28.ffn_down_exps.weight q5_0 +blk.28.ffn_gate_exps.weight q4_0 +blk.28.ffn_up_exps.weight q4_0 +blk.29.attn_k.weight q4_0 blk.29.attn_output.weight q4_K -blk.29.attn_v.weight q4_K -blk.29.ffn_down_exps.weight q4_K +blk.29.attn_q.weight q4_0 +blk.29.attn_v.weight q5_0 +blk.29.ffn_down_exps.weight q5_0 +blk.29.ffn_gate_exps.weight q4_0 +blk.29.ffn_up_exps.weight q4_0 +blk.30.attn_k.weight q4_0 blk.30.attn_output.weight q4_K -blk.30.attn_v.weight q4_K -blk.30.ffn_down_exps.weight q4_K +blk.30.attn_q.weight q4_0 +blk.30.attn_v.weight q5_0 +blk.30.ffn_down_exps.weight q5_0 +blk.30.ffn_gate_exps.weight q4_0 +blk.30.ffn_up_exps.weight q4_0 +blk.31.attn_k.weight q4_0 blk.31.attn_output.weight q4_K -blk.31.attn_v.weight q4_K -blk.31.ffn_down_exps.weight q4_K +blk.31.attn_q.weight q4_0 +blk.31.attn_v.weight q5_0 +blk.31.ffn_down_exps.weight q5_0 +blk.31.ffn_gate_exps.weight q4_0 +blk.31.ffn_up_exps.weight q4_0 +blk.32.attn_k.weight q4_0 blk.32.attn_output.weight q4_K -blk.32.attn_v.weight q4_K -blk.32.ffn_down_exps.weight q4_K +blk.32.attn_q.weight q4_0 +blk.32.attn_v.weight q5_0 +blk.32.ffn_down_exps.weight q5_0 +blk.32.ffn_gate_exps.weight q4_0 +blk.32.ffn_up_exps.weight q4_0 +blk.33.attn_k.weight q4_0 blk.33.attn_output.weight q4_K -blk.33.attn_v.weight q4_K -blk.33.ffn_down_exps.weight q4_K +blk.33.attn_q.weight q4_0 +blk.33.attn_v.weight q5_0 +blk.33.ffn_down_exps.weight q5_0 +blk.33.ffn_gate_exps.weight q4_0 +blk.33.ffn_up_exps.weight q4_0 +blk.34.attn_k.weight q4_0 blk.34.attn_output.weight q4_K -blk.34.attn_v.weight q4_K -blk.34.ffn_down_exps.weight q4_K +blk.34.attn_q.weight q4_0 +blk.34.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight q5_0 +blk.34.ffn_gate_exps.weight q4_0 +blk.34.ffn_up_exps.weight q4_0 +blk.35.attn_k.weight q4_0 blk.35.attn_output.weight q4_K -blk.35.attn_v.weight q4_K -blk.35.ffn_down_exps.weight q4_K +blk.35.attn_q.weight q4_0 +blk.35.attn_v.weight q5_0 +blk.35.ffn_down_exps.weight q5_0 +blk.35.ffn_gate_exps.weight q4_0 +blk.35.ffn_up_exps.weight q4_0 [Q3_K_L] q3_K output.weight q8_0 +token_embd.weight q4_0 +blk.0.attn_k.weight q4_0 blk.0.attn_output.weight q5_K -blk.0.attn_v.weight q5_K -blk.0.ffn_down_exps.weight q5_K +blk.0.attn_q.weight q4_0 +blk.0.attn_v.weight q5_1 +blk.0.ffn_down_exps.weight q5_1 +blk.0.ffn_gate_exps.weight q4_0 +blk.0.ffn_up_exps.weight q4_0 +blk.1.attn_k.weight q4_0 blk.1.attn_output.weight q5_K -blk.1.attn_v.weight q5_K -blk.1.ffn_down_exps.weight q5_K +blk.1.attn_q.weight q4_0 +blk.1.attn_v.weight q5_1 +blk.1.ffn_down_exps.weight q5_1 +blk.1.ffn_gate_exps.weight q4_0 +blk.1.ffn_up_exps.weight q4_0 +blk.2.attn_k.weight q4_0 blk.2.attn_output.weight q5_K -blk.2.attn_v.weight q5_K -blk.2.ffn_down_exps.weight q5_K +blk.2.attn_q.weight q4_0 +blk.2.attn_v.weight q5_1 +blk.2.ffn_down_exps.weight q5_1 +blk.2.ffn_gate_exps.weight q4_0 +blk.2.ffn_up_exps.weight q4_0 +blk.3.attn_k.weight q4_0 blk.3.attn_output.weight q5_K -blk.3.attn_v.weight q5_K -blk.3.ffn_down_exps.weight q5_K +blk.3.attn_q.weight q4_0 +blk.3.attn_v.weight q5_1 +blk.3.ffn_down_exps.weight q5_1 +blk.3.ffn_gate_exps.weight q4_0 +blk.3.ffn_up_exps.weight q4_0 +blk.4.attn_k.weight q4_0 blk.4.attn_output.weight q5_K -blk.4.attn_v.weight q5_K -blk.4.ffn_down_exps.weight q5_K +blk.4.attn_q.weight q4_0 +blk.4.attn_v.weight q5_1 +blk.4.ffn_down_exps.weight q5_1 +blk.4.ffn_gate_exps.weight q4_0 +blk.4.ffn_up_exps.weight q4_0 +blk.5.attn_k.weight q4_0 blk.5.attn_output.weight q5_K -blk.5.attn_v.weight q5_K -blk.5.ffn_down_exps.weight q5_K +blk.5.attn_q.weight q4_0 +blk.5.attn_v.weight q5_1 +blk.5.ffn_down_exps.weight q5_1 +blk.5.ffn_gate_exps.weight q4_0 +blk.5.ffn_up_exps.weight q4_0 +blk.6.attn_k.weight q4_0 blk.6.attn_output.weight q5_K -blk.6.attn_v.weight q5_K -blk.6.ffn_down_exps.weight q5_K +blk.6.attn_q.weight q4_0 +blk.6.attn_v.weight q5_1 +blk.6.ffn_down_exps.weight q5_1 +blk.6.ffn_gate_exps.weight q4_0 +blk.6.ffn_up_exps.weight q4_0 +blk.7.attn_k.weight q4_0 blk.7.attn_output.weight q5_K -blk.7.attn_v.weight q5_K -blk.7.ffn_down_exps.weight q5_K +blk.7.attn_q.weight q4_0 +blk.7.attn_v.weight q5_1 +blk.7.ffn_down_exps.weight q5_1 +blk.7.ffn_gate_exps.weight q4_0 +blk.7.ffn_up_exps.weight q4_0 +blk.8.attn_k.weight q4_0 blk.8.attn_output.weight q5_K -blk.8.attn_v.weight q5_K -blk.8.ffn_down_exps.weight q5_K +blk.8.attn_q.weight q4_0 +blk.8.attn_v.weight q5_1 +blk.8.ffn_down_exps.weight q5_1 +blk.8.ffn_gate_exps.weight q4_0 +blk.8.ffn_up_exps.weight q4_0 +blk.9.attn_k.weight q4_0 blk.9.attn_output.weight q5_K -blk.9.attn_v.weight q5_K -blk.9.ffn_down_exps.weight q5_K +blk.9.attn_q.weight q4_0 +blk.9.attn_v.weight q5_1 +blk.9.ffn_down_exps.weight q5_1 +blk.9.ffn_gate_exps.weight q4_0 +blk.9.ffn_up_exps.weight q4_0 +blk.10.attn_k.weight q4_0 blk.10.attn_output.weight q5_K -blk.10.attn_v.weight q5_K -blk.10.ffn_down_exps.weight q5_K +blk.10.attn_q.weight q4_0 +blk.10.attn_v.weight q5_1 +blk.10.ffn_down_exps.weight q5_1 +blk.10.ffn_gate_exps.weight q4_0 +blk.10.ffn_up_exps.weight q4_0 +blk.11.attn_k.weight q4_0 blk.11.attn_output.weight q5_K -blk.11.attn_v.weight q5_K -blk.11.ffn_down_exps.weight q5_K +blk.11.attn_q.weight q4_0 +blk.11.attn_v.weight q5_1 +blk.11.ffn_down_exps.weight q5_1 +blk.11.ffn_gate_exps.weight q4_0 +blk.11.ffn_up_exps.weight q4_0 +blk.12.attn_k.weight q4_0 blk.12.attn_output.weight q5_K -blk.12.attn_v.weight q5_K -blk.12.ffn_down_exps.weight q5_K +blk.12.attn_q.weight q4_0 +blk.12.attn_v.weight q5_1 +blk.12.ffn_down_exps.weight q5_1 +blk.12.ffn_gate_exps.weight q4_0 +blk.12.ffn_up_exps.weight q4_0 +blk.13.attn_k.weight q4_0 blk.13.attn_output.weight q5_K -blk.13.attn_v.weight q5_K -blk.13.ffn_down_exps.weight q5_K +blk.13.attn_q.weight q4_0 +blk.13.attn_v.weight q5_1 +blk.13.ffn_down_exps.weight q5_1 +blk.13.ffn_gate_exps.weight q4_0 +blk.13.ffn_up_exps.weight q4_0 +blk.14.attn_k.weight q4_0 blk.14.attn_output.weight q5_K -blk.14.attn_v.weight q5_K -blk.14.ffn_down_exps.weight q5_K +blk.14.attn_q.weight q4_0 +blk.14.attn_v.weight q5_1 +blk.14.ffn_down_exps.weight q5_1 +blk.14.ffn_gate_exps.weight q4_0 +blk.14.ffn_up_exps.weight q4_0 +blk.15.attn_k.weight q4_0 blk.15.attn_output.weight q5_K -blk.15.attn_v.weight q5_K -blk.15.ffn_down_exps.weight q5_K +blk.15.attn_q.weight q4_0 +blk.15.attn_v.weight q5_1 +blk.15.ffn_down_exps.weight q5_1 +blk.15.ffn_gate_exps.weight q4_0 +blk.15.ffn_up_exps.weight q4_0 +blk.16.attn_k.weight q4_0 blk.16.attn_output.weight q5_K -blk.16.attn_v.weight q5_K -blk.16.ffn_down_exps.weight q5_K +blk.16.attn_q.weight q4_0 +blk.16.attn_v.weight q5_1 +blk.16.ffn_down_exps.weight q5_1 +blk.16.ffn_gate_exps.weight q4_0 +blk.16.ffn_up_exps.weight q4_0 +blk.17.attn_k.weight q4_0 blk.17.attn_output.weight q5_K -blk.17.attn_v.weight q5_K -blk.17.ffn_down_exps.weight q5_K +blk.17.attn_q.weight q4_0 +blk.17.attn_v.weight q5_1 +blk.17.ffn_down_exps.weight q5_1 +blk.17.ffn_gate_exps.weight q4_0 +blk.17.ffn_up_exps.weight q4_0 +blk.18.attn_k.weight q4_0 blk.18.attn_output.weight q5_K -blk.18.attn_v.weight q5_K -blk.18.ffn_down_exps.weight q5_K +blk.18.attn_q.weight q4_0 +blk.18.attn_v.weight q5_1 +blk.18.ffn_down_exps.weight q5_1 +blk.18.ffn_gate_exps.weight q4_0 +blk.18.ffn_up_exps.weight q4_0 +blk.19.attn_k.weight q4_0 blk.19.attn_output.weight q5_K -blk.19.attn_v.weight q5_K -blk.19.ffn_down_exps.weight q5_K +blk.19.attn_q.weight q4_0 +blk.19.attn_v.weight q5_1 +blk.19.ffn_down_exps.weight q5_1 +blk.19.ffn_gate_exps.weight q4_0 +blk.19.ffn_up_exps.weight q4_0 +blk.20.attn_k.weight q4_0 blk.20.attn_output.weight q5_K -blk.20.attn_v.weight q5_K -blk.20.ffn_down_exps.weight q5_K +blk.20.attn_q.weight q4_0 +blk.20.attn_v.weight q5_1 +blk.20.ffn_down_exps.weight q5_1 +blk.20.ffn_gate_exps.weight q4_0 +blk.20.ffn_up_exps.weight q4_0 +blk.21.attn_k.weight q4_0 blk.21.attn_output.weight q5_K -blk.21.attn_v.weight q5_K -blk.21.ffn_down_exps.weight q5_K +blk.21.attn_q.weight q4_0 +blk.21.attn_v.weight q5_1 +blk.21.ffn_down_exps.weight q5_1 +blk.21.ffn_gate_exps.weight q4_0 +blk.21.ffn_up_exps.weight q4_0 +blk.22.attn_k.weight q4_0 blk.22.attn_output.weight q5_K -blk.22.attn_v.weight q5_K -blk.22.ffn_down_exps.weight q5_K +blk.22.attn_q.weight q4_0 +blk.22.attn_v.weight q5_1 +blk.22.ffn_down_exps.weight q5_1 +blk.22.ffn_gate_exps.weight q4_0 +blk.22.ffn_up_exps.weight q4_0 +blk.23.attn_k.weight q4_0 blk.23.attn_output.weight q5_K -blk.23.attn_v.weight q5_K -blk.23.ffn_down_exps.weight q5_K +blk.23.attn_q.weight q4_0 +blk.23.attn_v.weight q5_1 +blk.23.ffn_down_exps.weight q5_1 +blk.23.ffn_gate_exps.weight q4_0 +blk.23.ffn_up_exps.weight q4_0 +blk.24.attn_k.weight q4_0 blk.24.attn_output.weight q5_K -blk.24.attn_v.weight q5_K -blk.24.ffn_down_exps.weight q5_K +blk.24.attn_q.weight q4_0 +blk.24.attn_v.weight q5_1 +blk.24.ffn_down_exps.weight q5_1 +blk.24.ffn_gate_exps.weight q4_0 +blk.24.ffn_up_exps.weight q4_0 +blk.25.attn_k.weight q4_0 blk.25.attn_output.weight q5_K -blk.25.attn_v.weight q5_K -blk.25.ffn_down_exps.weight q5_K +blk.25.attn_q.weight q4_0 +blk.25.attn_v.weight q5_1 +blk.25.ffn_down_exps.weight q5_1 +blk.25.ffn_gate_exps.weight q4_0 +blk.25.ffn_up_exps.weight q4_0 +blk.26.attn_k.weight q4_0 blk.26.attn_output.weight q5_K -blk.26.attn_v.weight q5_K -blk.26.ffn_down_exps.weight q5_K +blk.26.attn_q.weight q4_0 +blk.26.attn_v.weight q5_1 +blk.26.ffn_down_exps.weight q5_1 +blk.26.ffn_gate_exps.weight q4_0 +blk.26.ffn_up_exps.weight q4_0 +blk.27.attn_k.weight q4_0 blk.27.attn_output.weight q5_K -blk.27.attn_v.weight q5_K -blk.27.ffn_down_exps.weight q5_K +blk.27.attn_q.weight q4_0 +blk.27.attn_v.weight q5_1 +blk.27.ffn_down_exps.weight q5_1 +blk.27.ffn_gate_exps.weight q4_0 +blk.27.ffn_up_exps.weight q4_0 +blk.28.attn_k.weight q4_0 blk.28.attn_output.weight q5_K -blk.28.attn_v.weight q5_K -blk.28.ffn_down_exps.weight q5_K +blk.28.attn_q.weight q4_0 +blk.28.attn_v.weight q5_1 +blk.28.ffn_down_exps.weight q5_1 +blk.28.ffn_gate_exps.weight q4_0 +blk.28.ffn_up_exps.weight q4_0 +blk.29.attn_k.weight q4_0 blk.29.attn_output.weight q5_K -blk.29.attn_v.weight q5_K -blk.29.ffn_down_exps.weight q5_K +blk.29.attn_q.weight q4_0 +blk.29.attn_v.weight q5_1 +blk.29.ffn_down_exps.weight q5_1 +blk.29.ffn_gate_exps.weight q4_0 +blk.29.ffn_up_exps.weight q4_0 +blk.30.attn_k.weight q4_0 blk.30.attn_output.weight q5_K -blk.30.attn_v.weight q5_K -blk.30.ffn_down_exps.weight q5_K +blk.30.attn_q.weight q4_0 +blk.30.attn_v.weight q5_1 +blk.30.ffn_down_exps.weight q5_1 +blk.30.ffn_gate_exps.weight q4_0 +blk.30.ffn_up_exps.weight q4_0 +blk.31.attn_k.weight q4_0 blk.31.attn_output.weight q5_K -blk.31.attn_v.weight q5_K -blk.31.ffn_down_exps.weight q5_K +blk.31.attn_q.weight q4_0 +blk.31.attn_v.weight q5_1 +blk.31.ffn_down_exps.weight q5_1 +blk.31.ffn_gate_exps.weight q4_0 +blk.31.ffn_up_exps.weight q4_0 +blk.32.attn_k.weight q4_0 blk.32.attn_output.weight q5_K -blk.32.attn_v.weight q5_K -blk.32.ffn_down_exps.weight q5_K +blk.32.attn_q.weight q4_0 +blk.32.attn_v.weight q5_1 +blk.32.ffn_down_exps.weight q5_1 +blk.32.ffn_gate_exps.weight q4_0 +blk.32.ffn_up_exps.weight q4_0 +blk.33.attn_k.weight q4_0 blk.33.attn_output.weight q5_K -blk.33.attn_v.weight q5_K -blk.33.ffn_down_exps.weight q5_K +blk.33.attn_q.weight q4_0 +blk.33.attn_v.weight q5_1 +blk.33.ffn_down_exps.weight q5_1 +blk.33.ffn_gate_exps.weight q4_0 +blk.33.ffn_up_exps.weight q4_0 +blk.34.attn_k.weight q4_0 blk.34.attn_output.weight q5_K -blk.34.attn_v.weight q5_K -blk.34.ffn_down_exps.weight q5_K +blk.34.attn_q.weight q4_0 +blk.34.attn_v.weight q5_1 +blk.34.ffn_down_exps.weight q5_1 +blk.34.ffn_gate_exps.weight q4_0 +blk.34.ffn_up_exps.weight q4_0 +blk.35.attn_k.weight q4_0 blk.35.attn_output.weight q5_K -blk.35.attn_v.weight q5_K -blk.35.ffn_down_exps.weight q5_K +blk.35.attn_q.weight q4_0 +blk.35.attn_v.weight q5_1 +blk.35.ffn_down_exps.weight q5_1 +blk.35.ffn_gate_exps.weight q4_0 +blk.35.ffn_up_exps.weight q4_0 [Q4_K_S] q4_K output.weight q8_0 -blk.0.attn_v.weight q5_K -blk.0.ffn_down_exps.weight q5_K -blk.1.attn_v.weight q5_K -blk.1.ffn_down_exps.weight q5_K -blk.2.attn_v.weight q5_K -blk.2.ffn_down_exps.weight q5_K -blk.3.attn_v.weight q5_K -blk.3.ffn_down_exps.weight q5_K +token_embd.weight q5_0 +blk.0.attn_k.weight q5_0 +blk.0.attn_q.weight q5_0 +blk.0.attn_v.weight q5_1 +blk.0.ffn_down_exps.weight q5_1 +blk.0.ffn_gate_exps.weight q5_0 +blk.0.ffn_up_exps.weight q5_0 +blk.1.attn_k.weight q5_0 +blk.1.attn_q.weight q5_0 +blk.1.attn_v.weight q5_1 +blk.1.ffn_down_exps.weight q5_1 +blk.1.ffn_gate_exps.weight q5_0 +blk.1.ffn_up_exps.weight q5_0 +blk.2.attn_k.weight q5_0 +blk.2.attn_q.weight q5_0 +blk.2.attn_v.weight q5_1 +blk.2.ffn_down_exps.weight q5_1 +blk.2.ffn_gate_exps.weight q5_0 +blk.2.ffn_up_exps.weight q5_0 +blk.3.attn_k.weight q5_0 +blk.3.attn_q.weight q5_0 +blk.3.attn_v.weight q5_1 +blk.3.ffn_down_exps.weight q5_1 +blk.3.ffn_gate_exps.weight q5_0 +blk.3.ffn_up_exps.weight q5_0 +blk.4.attn_k.weight q5_0 +blk.4.attn_q.weight q5_0 +blk.4.attn_v.weight q5_0 +blk.4.ffn_down_exps.weight q5_0 +blk.4.ffn_gate_exps.weight q5_0 +blk.4.ffn_up_exps.weight q5_0 +blk.5.attn_k.weight q5_0 +blk.5.attn_q.weight q5_0 +blk.5.attn_v.weight q5_0 +blk.5.ffn_down_exps.weight q5_0 +blk.5.ffn_gate_exps.weight q5_0 +blk.5.ffn_up_exps.weight q5_0 +blk.6.attn_k.weight q5_0 +blk.6.attn_q.weight q5_0 +blk.6.attn_v.weight q5_0 +blk.6.ffn_down_exps.weight q5_0 +blk.6.ffn_gate_exps.weight q5_0 +blk.6.ffn_up_exps.weight q5_0 +blk.7.attn_k.weight q5_0 +blk.7.attn_q.weight q5_0 +blk.7.attn_v.weight q5_0 +blk.7.ffn_down_exps.weight q5_0 +blk.7.ffn_gate_exps.weight q5_0 +blk.7.ffn_up_exps.weight q5_0 +blk.8.attn_k.weight q5_0 +blk.8.attn_q.weight q5_0 +blk.8.attn_v.weight q5_0 +blk.8.ffn_down_exps.weight q5_0 +blk.8.ffn_gate_exps.weight q5_0 +blk.8.ffn_up_exps.weight q5_0 +blk.9.attn_k.weight q5_0 +blk.9.attn_q.weight q5_0 +blk.9.attn_v.weight q5_0 +blk.9.ffn_down_exps.weight q5_0 +blk.9.ffn_gate_exps.weight q5_0 +blk.9.ffn_up_exps.weight q5_0 +blk.10.attn_k.weight q5_0 +blk.10.attn_q.weight q5_0 +blk.10.attn_v.weight q5_0 +blk.10.ffn_down_exps.weight q5_0 +blk.10.ffn_gate_exps.weight q5_0 +blk.10.ffn_up_exps.weight q5_0 +blk.11.attn_k.weight q5_0 +blk.11.attn_q.weight q5_0 +blk.11.attn_v.weight q5_0 +blk.11.ffn_down_exps.weight q5_0 +blk.11.ffn_gate_exps.weight q5_0 +blk.11.ffn_up_exps.weight q5_0 +blk.12.attn_k.weight q5_0 +blk.12.attn_q.weight q5_0 +blk.12.attn_v.weight q5_0 +blk.12.ffn_down_exps.weight q5_0 +blk.12.ffn_gate_exps.weight q5_0 +blk.12.ffn_up_exps.weight q5_0 +blk.13.attn_k.weight q5_0 +blk.13.attn_q.weight q5_0 +blk.13.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight q5_0 +blk.13.ffn_gate_exps.weight q5_0 +blk.13.ffn_up_exps.weight q5_0 +blk.14.attn_k.weight q5_0 +blk.14.attn_q.weight q5_0 +blk.14.attn_v.weight q5_0 +blk.14.ffn_down_exps.weight q5_0 +blk.14.ffn_gate_exps.weight q5_0 +blk.14.ffn_up_exps.weight q5_0 +blk.15.attn_k.weight q5_0 +blk.15.attn_q.weight q5_0 +blk.15.attn_v.weight q5_0 +blk.15.ffn_down_exps.weight q5_0 +blk.15.ffn_gate_exps.weight q5_0 +blk.15.ffn_up_exps.weight q5_0 +blk.16.attn_k.weight q5_0 +blk.16.attn_q.weight q5_0 +blk.16.attn_v.weight q5_0 +blk.16.ffn_down_exps.weight q5_0 +blk.16.ffn_gate_exps.weight q5_0 +blk.16.ffn_up_exps.weight q5_0 +blk.17.attn_k.weight q5_0 +blk.17.attn_q.weight q5_0 +blk.17.attn_v.weight q5_0 +blk.17.ffn_down_exps.weight q5_0 +blk.17.ffn_gate_exps.weight q5_0 +blk.17.ffn_up_exps.weight q5_0 +blk.18.attn_k.weight q5_0 +blk.18.attn_q.weight q5_0 +blk.18.attn_v.weight q5_0 +blk.18.ffn_down_exps.weight q5_0 +blk.18.ffn_gate_exps.weight q5_0 +blk.18.ffn_up_exps.weight q5_0 +blk.19.attn_k.weight q5_0 +blk.19.attn_q.weight q5_0 +blk.19.attn_v.weight q5_0 +blk.19.ffn_down_exps.weight q5_0 +blk.19.ffn_gate_exps.weight q5_0 +blk.19.ffn_up_exps.weight q5_0 +blk.20.attn_k.weight q5_0 +blk.20.attn_q.weight q5_0 +blk.20.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight q5_0 +blk.20.ffn_gate_exps.weight q5_0 +blk.20.ffn_up_exps.weight q5_0 +blk.21.attn_k.weight q5_0 +blk.21.attn_q.weight q5_0 +blk.21.attn_v.weight q5_0 +blk.21.ffn_down_exps.weight q5_0 +blk.21.ffn_gate_exps.weight q5_0 +blk.21.ffn_up_exps.weight q5_0 +blk.22.attn_k.weight q5_0 +blk.22.attn_q.weight q5_0 +blk.22.attn_v.weight q5_0 +blk.22.ffn_down_exps.weight q5_0 +blk.22.ffn_gate_exps.weight q5_0 +blk.22.ffn_up_exps.weight q5_0 +blk.23.attn_k.weight q5_0 +blk.23.attn_q.weight q5_0 +blk.23.attn_v.weight q5_0 +blk.23.ffn_down_exps.weight q5_0 +blk.23.ffn_gate_exps.weight q5_0 +blk.23.ffn_up_exps.weight q5_0 +blk.24.attn_k.weight q5_0 +blk.24.attn_q.weight q5_0 +blk.24.attn_v.weight q5_0 +blk.24.ffn_down_exps.weight q5_0 +blk.24.ffn_gate_exps.weight q5_0 +blk.24.ffn_up_exps.weight q5_0 +blk.25.attn_k.weight q5_0 +blk.25.attn_q.weight q5_0 +blk.25.attn_v.weight q5_0 +blk.25.ffn_down_exps.weight q5_0 +blk.25.ffn_gate_exps.weight q5_0 +blk.25.ffn_up_exps.weight q5_0 +blk.26.attn_k.weight q5_0 +blk.26.attn_q.weight q5_0 +blk.26.attn_v.weight q5_0 +blk.26.ffn_down_exps.weight q5_0 +blk.26.ffn_gate_exps.weight q5_0 +blk.26.ffn_up_exps.weight q5_0 +blk.27.attn_k.weight q5_0 +blk.27.attn_q.weight q5_0 +blk.27.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight q5_0 +blk.27.ffn_gate_exps.weight q5_0 +blk.27.ffn_up_exps.weight q5_0 +blk.28.attn_k.weight q5_0 +blk.28.attn_q.weight q5_0 +blk.28.attn_v.weight q5_0 +blk.28.ffn_down_exps.weight q5_0 +blk.28.ffn_gate_exps.weight q5_0 +blk.28.ffn_up_exps.weight q5_0 +blk.29.attn_k.weight q5_0 +blk.29.attn_q.weight q5_0 +blk.29.attn_v.weight q5_0 +blk.29.ffn_down_exps.weight q5_0 +blk.29.ffn_gate_exps.weight q5_0 +blk.29.ffn_up_exps.weight q5_0 +blk.30.attn_k.weight q5_0 +blk.30.attn_q.weight q5_0 +blk.30.attn_v.weight q5_0 +blk.30.ffn_down_exps.weight q5_0 +blk.30.ffn_gate_exps.weight q5_0 +blk.30.ffn_up_exps.weight q5_0 +blk.31.attn_k.weight q5_0 +blk.31.attn_q.weight q5_0 +blk.31.attn_v.weight q5_0 +blk.31.ffn_down_exps.weight q5_0 +blk.31.ffn_gate_exps.weight q5_0 +blk.31.ffn_up_exps.weight q5_0 +blk.32.attn_k.weight q5_0 +blk.32.attn_q.weight q5_0 +blk.32.attn_v.weight q5_0 +blk.32.ffn_down_exps.weight q5_0 +blk.32.ffn_gate_exps.weight q5_0 +blk.32.ffn_up_exps.weight q5_0 +blk.33.attn_k.weight q5_0 +blk.33.attn_q.weight q5_0 +blk.33.attn_v.weight q5_0 +blk.33.ffn_down_exps.weight q5_0 +blk.33.ffn_gate_exps.weight q5_0 +blk.33.ffn_up_exps.weight q5_0 +blk.34.attn_k.weight q5_0 +blk.34.attn_q.weight q5_0 +blk.34.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight q5_0 +blk.34.ffn_gate_exps.weight q5_0 +blk.34.ffn_up_exps.weight q5_0 +blk.35.attn_k.weight q5_0 +blk.35.attn_q.weight q5_0 +blk.35.attn_v.weight q5_0 +blk.35.ffn_down_exps.weight q5_0 +blk.35.ffn_gate_exps.weight q5_0 +blk.35.ffn_up_exps.weight q5_0 [Q4_K_M] q4_K output.weight q8_0 -blk.0.attn_v.weight q6_K -blk.0.ffn_down_exps.weight q6_K -blk.1.attn_v.weight q6_K -blk.1.ffn_down_exps.weight q6_K -blk.2.attn_v.weight q6_K -blk.2.ffn_down_exps.weight q6_K -blk.3.attn_v.weight q6_K -blk.3.ffn_down_exps.weight q6_K -blk.6.attn_v.weight q6_K -blk.6.ffn_down_exps.weight q6_K -blk.9.attn_v.weight q6_K -blk.9.ffn_down_exps.weight q6_K -blk.12.attn_v.weight q6_K -blk.12.ffn_down_exps.weight q6_K -blk.15.attn_v.weight q6_K -blk.15.ffn_down_exps.weight q6_K -blk.18.attn_v.weight q6_K -blk.18.ffn_down_exps.weight q6_K -blk.21.attn_v.weight q6_K -blk.21.ffn_down_exps.weight q6_K -blk.24.attn_v.weight q6_K -blk.24.ffn_down_exps.weight q6_K -blk.27.attn_v.weight q6_K -blk.27.ffn_down_exps.weight q6_K -blk.30.attn_v.weight q6_K -blk.30.ffn_down_exps.weight q6_K -blk.31.attn_v.weight q6_K -blk.31.ffn_down_exps.weight q6_K -blk.32.attn_v.weight q6_K -blk.32.ffn_down_exps.weight q6_K -blk.33.attn_v.weight q6_K -blk.33.ffn_down_exps.weight q6_K -blk.34.attn_v.weight q6_K -blk.34.ffn_down_exps.weight q6_K -blk.35.attn_v.weight q6_K -blk.35.ffn_down_exps.weight q6_K +token_embd.weight q5_0 +blk.0.attn_k.weight q5_0 +blk.0.attn_q.weight q5_0 +blk.0.attn_v.weight q8_0 +blk.0.ffn_down_exps.weight q8_0 +blk.0.ffn_gate_exps.weight q5_0 +blk.0.ffn_up_exps.weight q5_0 +blk.1.attn_k.weight q5_0 +blk.1.attn_q.weight q5_0 +blk.1.attn_v.weight q8_0 +blk.1.ffn_down_exps.weight q8_0 +blk.1.ffn_gate_exps.weight q5_0 +blk.1.ffn_up_exps.weight q5_0 +blk.2.attn_k.weight q5_0 +blk.2.attn_q.weight q5_0 +blk.2.attn_v.weight q8_0 +blk.2.ffn_down_exps.weight q8_0 +blk.2.ffn_gate_exps.weight q5_0 +blk.2.ffn_up_exps.weight q5_0 +blk.3.attn_k.weight q5_0 +blk.3.attn_q.weight q5_0 +blk.3.attn_v.weight q8_0 +blk.3.ffn_down_exps.weight q8_0 +blk.3.ffn_gate_exps.weight q5_0 +blk.3.ffn_up_exps.weight q5_0 +blk.4.attn_k.weight q5_0 +blk.4.attn_q.weight q5_0 +blk.4.attn_v.weight q5_0 +blk.4.ffn_down_exps.weight q5_0 +blk.4.ffn_gate_exps.weight q5_0 +blk.4.ffn_up_exps.weight q5_0 +blk.5.attn_k.weight q5_0 +blk.5.attn_q.weight q5_0 +blk.5.attn_v.weight q5_0 +blk.5.ffn_down_exps.weight q5_0 +blk.5.ffn_gate_exps.weight q5_0 +blk.5.ffn_up_exps.weight q5_0 +blk.6.attn_k.weight q5_0 +blk.6.attn_q.weight q5_0 +blk.6.attn_v.weight q8_0 +blk.6.ffn_down_exps.weight q8_0 +blk.6.ffn_gate_exps.weight q5_0 +blk.6.ffn_up_exps.weight q5_0 +blk.7.attn_k.weight q5_0 +blk.7.attn_q.weight q5_0 +blk.7.attn_v.weight q5_0 +blk.7.ffn_down_exps.weight q5_0 +blk.7.ffn_gate_exps.weight q5_0 +blk.7.ffn_up_exps.weight q5_0 +blk.8.attn_k.weight q5_0 +blk.8.attn_q.weight q5_0 +blk.8.attn_v.weight q5_0 +blk.8.ffn_down_exps.weight q5_0 +blk.8.ffn_gate_exps.weight q5_0 +blk.8.ffn_up_exps.weight q5_0 +blk.9.attn_k.weight q5_0 +blk.9.attn_q.weight q5_0 +blk.9.attn_v.weight q8_0 +blk.9.ffn_down_exps.weight q8_0 +blk.9.ffn_gate_exps.weight q5_0 +blk.9.ffn_up_exps.weight q5_0 +blk.10.attn_k.weight q5_0 +blk.10.attn_q.weight q5_0 +blk.10.attn_v.weight q5_0 +blk.10.ffn_down_exps.weight q5_0 +blk.10.ffn_gate_exps.weight q5_0 +blk.10.ffn_up_exps.weight q5_0 +blk.11.attn_k.weight q5_0 +blk.11.attn_q.weight q5_0 +blk.11.attn_v.weight q5_0 +blk.11.ffn_down_exps.weight q5_0 +blk.11.ffn_gate_exps.weight q5_0 +blk.11.ffn_up_exps.weight q5_0 +blk.12.attn_k.weight q5_0 +blk.12.attn_q.weight q5_0 +blk.12.attn_v.weight q8_0 +blk.12.ffn_down_exps.weight q8_0 +blk.12.ffn_gate_exps.weight q5_0 +blk.12.ffn_up_exps.weight q5_0 +blk.13.attn_k.weight q5_0 +blk.13.attn_q.weight q5_0 +blk.13.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight q5_0 +blk.13.ffn_gate_exps.weight q5_0 +blk.13.ffn_up_exps.weight q5_0 +blk.14.attn_k.weight q5_0 +blk.14.attn_q.weight q5_0 +blk.14.attn_v.weight q5_0 +blk.14.ffn_down_exps.weight q5_0 +blk.14.ffn_gate_exps.weight q5_0 +blk.14.ffn_up_exps.weight q5_0 +blk.15.attn_k.weight q5_0 +blk.15.attn_q.weight q5_0 +blk.15.attn_v.weight q8_0 +blk.15.ffn_down_exps.weight q8_0 +blk.15.ffn_gate_exps.weight q5_0 +blk.15.ffn_up_exps.weight q5_0 +blk.16.attn_k.weight q5_0 +blk.16.attn_q.weight q5_0 +blk.16.attn_v.weight q5_0 +blk.16.ffn_down_exps.weight q5_0 +blk.16.ffn_gate_exps.weight q5_0 +blk.16.ffn_up_exps.weight q5_0 +blk.17.attn_k.weight q5_0 +blk.17.attn_q.weight q5_0 +blk.17.attn_v.weight q5_0 +blk.17.ffn_down_exps.weight q5_0 +blk.17.ffn_gate_exps.weight q5_0 +blk.17.ffn_up_exps.weight q5_0 +blk.18.attn_k.weight q5_0 +blk.18.attn_q.weight q5_0 +blk.18.attn_v.weight q8_0 +blk.18.ffn_down_exps.weight q8_0 +blk.18.ffn_gate_exps.weight q5_0 +blk.18.ffn_up_exps.weight q5_0 +blk.19.attn_k.weight q5_0 +blk.19.attn_q.weight q5_0 +blk.19.attn_v.weight q5_0 +blk.19.ffn_down_exps.weight q5_0 +blk.19.ffn_gate_exps.weight q5_0 +blk.19.ffn_up_exps.weight q5_0 +blk.20.attn_k.weight q5_0 +blk.20.attn_q.weight q5_0 +blk.20.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight q5_0 +blk.20.ffn_gate_exps.weight q5_0 +blk.20.ffn_up_exps.weight q5_0 +blk.21.attn_k.weight q5_0 +blk.21.attn_q.weight q5_0 +blk.21.attn_v.weight q8_0 +blk.21.ffn_down_exps.weight q8_0 +blk.21.ffn_gate_exps.weight q5_0 +blk.21.ffn_up_exps.weight q5_0 +blk.22.attn_k.weight q5_0 +blk.22.attn_q.weight q5_0 +blk.22.attn_v.weight q5_0 +blk.22.ffn_down_exps.weight q5_0 +blk.22.ffn_gate_exps.weight q5_0 +blk.22.ffn_up_exps.weight q5_0 +blk.23.attn_k.weight q5_0 +blk.23.attn_q.weight q5_0 +blk.23.attn_v.weight q5_0 +blk.23.ffn_down_exps.weight q5_0 +blk.23.ffn_gate_exps.weight q5_0 +blk.23.ffn_up_exps.weight q5_0 +blk.24.attn_k.weight q5_0 +blk.24.attn_q.weight q5_0 +blk.24.attn_v.weight q8_0 +blk.24.ffn_down_exps.weight q8_0 +blk.24.ffn_gate_exps.weight q5_0 +blk.24.ffn_up_exps.weight q5_0 +blk.25.attn_k.weight q5_0 +blk.25.attn_q.weight q5_0 +blk.25.attn_v.weight q5_0 +blk.25.ffn_down_exps.weight q5_0 +blk.25.ffn_gate_exps.weight q5_0 +blk.25.ffn_up_exps.weight q5_0 +blk.26.attn_k.weight q5_0 +blk.26.attn_q.weight q5_0 +blk.26.attn_v.weight q5_0 +blk.26.ffn_down_exps.weight q5_0 +blk.26.ffn_gate_exps.weight q5_0 +blk.26.ffn_up_exps.weight q5_0 +blk.27.attn_k.weight q5_0 +blk.27.attn_q.weight q5_0 +blk.27.attn_v.weight q8_0 +blk.27.ffn_down_exps.weight q8_0 +blk.27.ffn_gate_exps.weight q5_0 +blk.27.ffn_up_exps.weight q5_0 +blk.28.attn_k.weight q5_0 +blk.28.attn_q.weight q5_0 +blk.28.attn_v.weight q5_0 +blk.28.ffn_down_exps.weight q5_0 +blk.28.ffn_gate_exps.weight q5_0 +blk.28.ffn_up_exps.weight q5_0 +blk.29.attn_k.weight q5_0 +blk.29.attn_q.weight q5_0 +blk.29.attn_v.weight q5_0 +blk.29.ffn_down_exps.weight q5_0 +blk.29.ffn_gate_exps.weight q5_0 +blk.29.ffn_up_exps.weight q5_0 +blk.30.attn_k.weight q5_0 +blk.30.attn_q.weight q5_0 +blk.30.attn_v.weight q8_0 +blk.30.ffn_down_exps.weight q8_0 +blk.30.ffn_gate_exps.weight q5_0 +blk.30.ffn_up_exps.weight q5_0 +blk.31.attn_k.weight q5_0 +blk.31.attn_q.weight q5_0 +blk.31.attn_v.weight q8_0 +blk.31.ffn_down_exps.weight q8_0 +blk.31.ffn_gate_exps.weight q5_0 +blk.31.ffn_up_exps.weight q5_0 +blk.32.attn_k.weight q5_0 +blk.32.attn_q.weight q5_0 +blk.32.attn_v.weight q8_0 +blk.32.ffn_down_exps.weight q8_0 +blk.32.ffn_gate_exps.weight q5_0 +blk.32.ffn_up_exps.weight q5_0 +blk.33.attn_k.weight q5_0 +blk.33.attn_q.weight q5_0 +blk.33.attn_v.weight q8_0 +blk.33.ffn_down_exps.weight q8_0 +blk.33.ffn_gate_exps.weight q5_0 +blk.33.ffn_up_exps.weight q5_0 +blk.34.attn_k.weight q5_0 +blk.34.attn_q.weight q5_0 +blk.34.attn_v.weight q8_0 +blk.34.ffn_down_exps.weight q8_0 +blk.34.ffn_gate_exps.weight q5_0 +blk.34.ffn_up_exps.weight q5_0 +blk.35.attn_k.weight q5_0 +blk.35.attn_q.weight q5_0 +blk.35.attn_v.weight q8_0 +blk.35.ffn_down_exps.weight q8_0 +blk.35.ffn_gate_exps.weight q5_0 +blk.35.ffn_up_exps.weight q5_0 [Q5_K_S] q5_K output.weight q8_0 +token_embd.weight q5_1 +blk.0.attn_k.weight q5_1 +blk.0.attn_q.weight q5_1 +blk.0.attn_v.weight q5_1 +blk.0.ffn_down_exps.weight q5_1 +blk.0.ffn_gate_exps.weight q5_1 +blk.0.ffn_up_exps.weight q5_1 +blk.1.attn_k.weight q5_1 +blk.1.attn_q.weight q5_1 +blk.1.attn_v.weight q5_1 +blk.1.ffn_down_exps.weight q5_1 +blk.1.ffn_gate_exps.weight q5_1 +blk.1.ffn_up_exps.weight q5_1 +blk.2.attn_k.weight q5_1 +blk.2.attn_q.weight q5_1 +blk.2.attn_v.weight q5_1 +blk.2.ffn_down_exps.weight q5_1 +blk.2.ffn_gate_exps.weight q5_1 +blk.2.ffn_up_exps.weight q5_1 +blk.3.attn_k.weight q5_1 +blk.3.attn_q.weight q5_1 +blk.3.attn_v.weight q5_1 +blk.3.ffn_down_exps.weight q5_1 +blk.3.ffn_gate_exps.weight q5_1 +blk.3.ffn_up_exps.weight q5_1 +blk.4.attn_k.weight q5_1 +blk.4.attn_q.weight q5_1 +blk.4.attn_v.weight q5_1 +blk.4.ffn_down_exps.weight q5_1 +blk.4.ffn_gate_exps.weight q5_1 +blk.4.ffn_up_exps.weight q5_1 +blk.5.attn_k.weight q5_1 +blk.5.attn_q.weight q5_1 +blk.5.attn_v.weight q5_1 +blk.5.ffn_down_exps.weight q5_1 +blk.5.ffn_gate_exps.weight q5_1 +blk.5.ffn_up_exps.weight q5_1 +blk.6.attn_k.weight q5_1 +blk.6.attn_q.weight q5_1 +blk.6.attn_v.weight q5_1 +blk.6.ffn_down_exps.weight q5_1 +blk.6.ffn_gate_exps.weight q5_1 +blk.6.ffn_up_exps.weight q5_1 +blk.7.attn_k.weight q5_1 +blk.7.attn_q.weight q5_1 +blk.7.attn_v.weight q5_1 +blk.7.ffn_down_exps.weight q5_1 +blk.7.ffn_gate_exps.weight q5_1 +blk.7.ffn_up_exps.weight q5_1 +blk.8.attn_k.weight q5_1 +blk.8.attn_q.weight q5_1 +blk.8.attn_v.weight q5_1 +blk.8.ffn_down_exps.weight q5_1 +blk.8.ffn_gate_exps.weight q5_1 +blk.8.ffn_up_exps.weight q5_1 +blk.9.attn_k.weight q5_1 +blk.9.attn_q.weight q5_1 +blk.9.attn_v.weight q5_1 +blk.9.ffn_down_exps.weight q5_1 +blk.9.ffn_gate_exps.weight q5_1 +blk.9.ffn_up_exps.weight q5_1 +blk.10.attn_k.weight q5_1 +blk.10.attn_q.weight q5_1 +blk.10.attn_v.weight q5_1 +blk.10.ffn_down_exps.weight q5_1 +blk.10.ffn_gate_exps.weight q5_1 +blk.10.ffn_up_exps.weight q5_1 +blk.11.attn_k.weight q5_1 +blk.11.attn_q.weight q5_1 +blk.11.attn_v.weight q5_1 +blk.11.ffn_down_exps.weight q5_1 +blk.11.ffn_gate_exps.weight q5_1 +blk.11.ffn_up_exps.weight q5_1 +blk.12.attn_k.weight q5_1 +blk.12.attn_q.weight q5_1 +blk.12.attn_v.weight q5_1 +blk.12.ffn_down_exps.weight q5_1 +blk.12.ffn_gate_exps.weight q5_1 +blk.12.ffn_up_exps.weight q5_1 +blk.13.attn_k.weight q5_1 +blk.13.attn_q.weight q5_1 +blk.13.attn_v.weight q5_1 +blk.13.ffn_down_exps.weight q5_1 +blk.13.ffn_gate_exps.weight q5_1 +blk.13.ffn_up_exps.weight q5_1 +blk.14.attn_k.weight q5_1 +blk.14.attn_q.weight q5_1 +blk.14.attn_v.weight q5_1 +blk.14.ffn_down_exps.weight q5_1 +blk.14.ffn_gate_exps.weight q5_1 +blk.14.ffn_up_exps.weight q5_1 +blk.15.attn_k.weight q5_1 +blk.15.attn_q.weight q5_1 +blk.15.attn_v.weight q5_1 +blk.15.ffn_down_exps.weight q5_1 +blk.15.ffn_gate_exps.weight q5_1 +blk.15.ffn_up_exps.weight q5_1 +blk.16.attn_k.weight q5_1 +blk.16.attn_q.weight q5_1 +blk.16.attn_v.weight q5_1 +blk.16.ffn_down_exps.weight q5_1 +blk.16.ffn_gate_exps.weight q5_1 +blk.16.ffn_up_exps.weight q5_1 +blk.17.attn_k.weight q5_1 +blk.17.attn_q.weight q5_1 +blk.17.attn_v.weight q5_1 +blk.17.ffn_down_exps.weight q5_1 +blk.17.ffn_gate_exps.weight q5_1 +blk.17.ffn_up_exps.weight q5_1 +blk.18.attn_k.weight q5_1 +blk.18.attn_q.weight q5_1 +blk.18.attn_v.weight q5_1 +blk.18.ffn_down_exps.weight q5_1 +blk.18.ffn_gate_exps.weight q5_1 +blk.18.ffn_up_exps.weight q5_1 +blk.19.attn_k.weight q5_1 +blk.19.attn_q.weight q5_1 +blk.19.attn_v.weight q5_1 +blk.19.ffn_down_exps.weight q5_1 +blk.19.ffn_gate_exps.weight q5_1 +blk.19.ffn_up_exps.weight q5_1 +blk.20.attn_k.weight q5_1 +blk.20.attn_q.weight q5_1 +blk.20.attn_v.weight q5_1 +blk.20.ffn_down_exps.weight q5_1 +blk.20.ffn_gate_exps.weight q5_1 +blk.20.ffn_up_exps.weight q5_1 +blk.21.attn_k.weight q5_1 +blk.21.attn_q.weight q5_1 +blk.21.attn_v.weight q5_1 +blk.21.ffn_down_exps.weight q5_1 +blk.21.ffn_gate_exps.weight q5_1 +blk.21.ffn_up_exps.weight q5_1 +blk.22.attn_k.weight q5_1 +blk.22.attn_q.weight q5_1 +blk.22.attn_v.weight q5_1 +blk.22.ffn_down_exps.weight q5_1 +blk.22.ffn_gate_exps.weight q5_1 +blk.22.ffn_up_exps.weight q5_1 +blk.23.attn_k.weight q5_1 +blk.23.attn_q.weight q5_1 +blk.23.attn_v.weight q5_1 +blk.23.ffn_down_exps.weight q5_1 +blk.23.ffn_gate_exps.weight q5_1 +blk.23.ffn_up_exps.weight q5_1 +blk.24.attn_k.weight q5_1 +blk.24.attn_q.weight q5_1 +blk.24.attn_v.weight q5_1 +blk.24.ffn_down_exps.weight q5_1 +blk.24.ffn_gate_exps.weight q5_1 +blk.24.ffn_up_exps.weight q5_1 +blk.25.attn_k.weight q5_1 +blk.25.attn_q.weight q5_1 +blk.25.attn_v.weight q5_1 +blk.25.ffn_down_exps.weight q5_1 +blk.25.ffn_gate_exps.weight q5_1 +blk.25.ffn_up_exps.weight q5_1 +blk.26.attn_k.weight q5_1 +blk.26.attn_q.weight q5_1 +blk.26.attn_v.weight q5_1 +blk.26.ffn_down_exps.weight q5_1 +blk.26.ffn_gate_exps.weight q5_1 +blk.26.ffn_up_exps.weight q5_1 +blk.27.attn_k.weight q5_1 +blk.27.attn_q.weight q5_1 +blk.27.attn_v.weight q5_1 +blk.27.ffn_down_exps.weight q5_1 +blk.27.ffn_gate_exps.weight q5_1 +blk.27.ffn_up_exps.weight q5_1 +blk.28.attn_k.weight q5_1 +blk.28.attn_q.weight q5_1 +blk.28.attn_v.weight q5_1 +blk.28.ffn_down_exps.weight q5_1 +blk.28.ffn_gate_exps.weight q5_1 +blk.28.ffn_up_exps.weight q5_1 +blk.29.attn_k.weight q5_1 +blk.29.attn_q.weight q5_1 +blk.29.attn_v.weight q5_1 +blk.29.ffn_down_exps.weight q5_1 +blk.29.ffn_gate_exps.weight q5_1 +blk.29.ffn_up_exps.weight q5_1 +blk.30.attn_k.weight q5_1 +blk.30.attn_q.weight q5_1 +blk.30.attn_v.weight q5_1 +blk.30.ffn_down_exps.weight q5_1 +blk.30.ffn_gate_exps.weight q5_1 +blk.30.ffn_up_exps.weight q5_1 +blk.31.attn_k.weight q5_1 +blk.31.attn_q.weight q5_1 +blk.31.attn_v.weight q5_1 +blk.31.ffn_down_exps.weight q5_1 +blk.31.ffn_gate_exps.weight q5_1 +blk.31.ffn_up_exps.weight q5_1 +blk.32.attn_k.weight q5_1 +blk.32.attn_q.weight q5_1 +blk.32.attn_v.weight q5_1 +blk.32.ffn_down_exps.weight q5_1 +blk.32.ffn_gate_exps.weight q5_1 +blk.32.ffn_up_exps.weight q5_1 +blk.33.attn_k.weight q5_1 +blk.33.attn_q.weight q5_1 +blk.33.attn_v.weight q5_1 +blk.33.ffn_down_exps.weight q5_1 +blk.33.ffn_gate_exps.weight q5_1 +blk.33.ffn_up_exps.weight q5_1 +blk.34.attn_k.weight q5_1 +blk.34.attn_q.weight q5_1 +blk.34.attn_v.weight q5_1 +blk.34.ffn_down_exps.weight q5_1 +blk.34.ffn_gate_exps.weight q5_1 +blk.34.ffn_up_exps.weight q5_1 +blk.35.attn_k.weight q5_1 +blk.35.attn_q.weight q5_1 +blk.35.attn_v.weight q5_1 +blk.35.ffn_down_exps.weight q5_1 +blk.35.ffn_gate_exps.weight q5_1 +blk.35.ffn_up_exps.weight q5_1 [Q5_K_M] q5_K output.weight q8_0 -blk.0.attn_v.weight q6_K -blk.0.ffn_down_exps.weight q6_K -blk.1.attn_v.weight q6_K -blk.1.ffn_down_exps.weight q6_K -blk.2.attn_v.weight q6_K -blk.2.ffn_down_exps.weight q6_K -blk.3.attn_v.weight q6_K -blk.3.ffn_down_exps.weight q6_K -blk.6.attn_v.weight q6_K -blk.6.ffn_down_exps.weight q6_K -blk.9.attn_v.weight q6_K -blk.9.ffn_down_exps.weight q6_K -blk.12.attn_v.weight q6_K -blk.12.ffn_down_exps.weight q6_K -blk.15.attn_v.weight q6_K -blk.15.ffn_down_exps.weight q6_K -blk.18.attn_v.weight q6_K -blk.18.ffn_down_exps.weight q6_K -blk.21.attn_v.weight q6_K -blk.21.ffn_down_exps.weight q6_K -blk.24.attn_v.weight q6_K -blk.24.ffn_down_exps.weight q6_K -blk.27.attn_v.weight q6_K -blk.27.ffn_down_exps.weight q6_K -blk.30.attn_v.weight q6_K -blk.30.ffn_down_exps.weight q6_K -blk.31.attn_v.weight q6_K -blk.31.ffn_down_exps.weight q6_K -blk.32.attn_v.weight q6_K -blk.32.ffn_down_exps.weight q6_K -blk.33.attn_v.weight q6_K -blk.33.ffn_down_exps.weight q6_K -blk.34.attn_v.weight q6_K -blk.34.ffn_down_exps.weight q6_K -blk.35.attn_v.weight q6_K -blk.35.ffn_down_exps.weight q6_K +token_embd.weight q5_1 +blk.0.attn_k.weight q5_1 +blk.0.attn_q.weight q5_1 +blk.0.attn_v.weight q8_0 +blk.0.ffn_down_exps.weight q8_0 +blk.0.ffn_gate_exps.weight q5_1 +blk.0.ffn_up_exps.weight q5_1 +blk.1.attn_k.weight q5_1 +blk.1.attn_q.weight q5_1 +blk.1.attn_v.weight q8_0 +blk.1.ffn_down_exps.weight q8_0 +blk.1.ffn_gate_exps.weight q5_1 +blk.1.ffn_up_exps.weight q5_1 +blk.2.attn_k.weight q5_1 +blk.2.attn_q.weight q5_1 +blk.2.attn_v.weight q8_0 +blk.2.ffn_down_exps.weight q8_0 +blk.2.ffn_gate_exps.weight q5_1 +blk.2.ffn_up_exps.weight q5_1 +blk.3.attn_k.weight q5_1 +blk.3.attn_q.weight q5_1 +blk.3.attn_v.weight q8_0 +blk.3.ffn_down_exps.weight q8_0 +blk.3.ffn_gate_exps.weight q5_1 +blk.3.ffn_up_exps.weight q5_1 +blk.4.attn_k.weight q5_1 +blk.4.attn_q.weight q5_1 +blk.4.attn_v.weight q5_1 +blk.4.ffn_down_exps.weight q5_1 +blk.4.ffn_gate_exps.weight q5_1 +blk.4.ffn_up_exps.weight q5_1 +blk.5.attn_k.weight q5_1 +blk.5.attn_q.weight q5_1 +blk.5.attn_v.weight q5_1 +blk.5.ffn_down_exps.weight q5_1 +blk.5.ffn_gate_exps.weight q5_1 +blk.5.ffn_up_exps.weight q5_1 +blk.6.attn_k.weight q5_1 +blk.6.attn_q.weight q5_1 +blk.6.attn_v.weight q8_0 +blk.6.ffn_down_exps.weight q8_0 +blk.6.ffn_gate_exps.weight q5_1 +blk.6.ffn_up_exps.weight q5_1 +blk.7.attn_k.weight q5_1 +blk.7.attn_q.weight q5_1 +blk.7.attn_v.weight q5_1 +blk.7.ffn_down_exps.weight q5_1 +blk.7.ffn_gate_exps.weight q5_1 +blk.7.ffn_up_exps.weight q5_1 +blk.8.attn_k.weight q5_1 +blk.8.attn_q.weight q5_1 +blk.8.attn_v.weight q5_1 +blk.8.ffn_down_exps.weight q5_1 +blk.8.ffn_gate_exps.weight q5_1 +blk.8.ffn_up_exps.weight q5_1 +blk.9.attn_k.weight q5_1 +blk.9.attn_q.weight q5_1 +blk.9.attn_v.weight q8_0 +blk.9.ffn_down_exps.weight q8_0 +blk.9.ffn_gate_exps.weight q5_1 +blk.9.ffn_up_exps.weight q5_1 +blk.10.attn_k.weight q5_1 +blk.10.attn_q.weight q5_1 +blk.10.attn_v.weight q5_1 +blk.10.ffn_down_exps.weight q5_1 +blk.10.ffn_gate_exps.weight q5_1 +blk.10.ffn_up_exps.weight q5_1 +blk.11.attn_k.weight q5_1 +blk.11.attn_q.weight q5_1 +blk.11.attn_v.weight q5_1 +blk.11.ffn_down_exps.weight q5_1 +blk.11.ffn_gate_exps.weight q5_1 +blk.11.ffn_up_exps.weight q5_1 +blk.12.attn_k.weight q5_1 +blk.12.attn_q.weight q5_1 +blk.12.attn_v.weight q8_0 +blk.12.ffn_down_exps.weight q8_0 +blk.12.ffn_gate_exps.weight q5_1 +blk.12.ffn_up_exps.weight q5_1 +blk.13.attn_k.weight q5_1 +blk.13.attn_q.weight q5_1 +blk.13.attn_v.weight q5_1 +blk.13.ffn_down_exps.weight q5_1 +blk.13.ffn_gate_exps.weight q5_1 +blk.13.ffn_up_exps.weight q5_1 +blk.14.attn_k.weight q5_1 +blk.14.attn_q.weight q5_1 +blk.14.attn_v.weight q5_1 +blk.14.ffn_down_exps.weight q5_1 +blk.14.ffn_gate_exps.weight q5_1 +blk.14.ffn_up_exps.weight q5_1 +blk.15.attn_k.weight q5_1 +blk.15.attn_q.weight q5_1 +blk.15.attn_v.weight q8_0 +blk.15.ffn_down_exps.weight q8_0 +blk.15.ffn_gate_exps.weight q5_1 +blk.15.ffn_up_exps.weight q5_1 +blk.16.attn_k.weight q5_1 +blk.16.attn_q.weight q5_1 +blk.16.attn_v.weight q5_1 +blk.16.ffn_down_exps.weight q5_1 +blk.16.ffn_gate_exps.weight q5_1 +blk.16.ffn_up_exps.weight q5_1 +blk.17.attn_k.weight q5_1 +blk.17.attn_q.weight q5_1 +blk.17.attn_v.weight q5_1 +blk.17.ffn_down_exps.weight q5_1 +blk.17.ffn_gate_exps.weight q5_1 +blk.17.ffn_up_exps.weight q5_1 +blk.18.attn_k.weight q5_1 +blk.18.attn_q.weight q5_1 +blk.18.attn_v.weight q8_0 +blk.18.ffn_down_exps.weight q8_0 +blk.18.ffn_gate_exps.weight q5_1 +blk.18.ffn_up_exps.weight q5_1 +blk.19.attn_k.weight q5_1 +blk.19.attn_q.weight q5_1 +blk.19.attn_v.weight q5_1 +blk.19.ffn_down_exps.weight q5_1 +blk.19.ffn_gate_exps.weight q5_1 +blk.19.ffn_up_exps.weight q5_1 +blk.20.attn_k.weight q5_1 +blk.20.attn_q.weight q5_1 +blk.20.attn_v.weight q5_1 +blk.20.ffn_down_exps.weight q5_1 +blk.20.ffn_gate_exps.weight q5_1 +blk.20.ffn_up_exps.weight q5_1 +blk.21.attn_k.weight q5_1 +blk.21.attn_q.weight q5_1 +blk.21.attn_v.weight q8_0 +blk.21.ffn_down_exps.weight q8_0 +blk.21.ffn_gate_exps.weight q5_1 +blk.21.ffn_up_exps.weight q5_1 +blk.22.attn_k.weight q5_1 +blk.22.attn_q.weight q5_1 +blk.22.attn_v.weight q5_1 +blk.22.ffn_down_exps.weight q5_1 +blk.22.ffn_gate_exps.weight q5_1 +blk.22.ffn_up_exps.weight q5_1 +blk.23.attn_k.weight q5_1 +blk.23.attn_q.weight q5_1 +blk.23.attn_v.weight q5_1 +blk.23.ffn_down_exps.weight q5_1 +blk.23.ffn_gate_exps.weight q5_1 +blk.23.ffn_up_exps.weight q5_1 +blk.24.attn_k.weight q5_1 +blk.24.attn_q.weight q5_1 +blk.24.attn_v.weight q8_0 +blk.24.ffn_down_exps.weight q8_0 +blk.24.ffn_gate_exps.weight q5_1 +blk.24.ffn_up_exps.weight q5_1 +blk.25.attn_k.weight q5_1 +blk.25.attn_q.weight q5_1 +blk.25.attn_v.weight q5_1 +blk.25.ffn_down_exps.weight q5_1 +blk.25.ffn_gate_exps.weight q5_1 +blk.25.ffn_up_exps.weight q5_1 +blk.26.attn_k.weight q5_1 +blk.26.attn_q.weight q5_1 +blk.26.attn_v.weight q5_1 +blk.26.ffn_down_exps.weight q5_1 +blk.26.ffn_gate_exps.weight q5_1 +blk.26.ffn_up_exps.weight q5_1 +blk.27.attn_k.weight q5_1 +blk.27.attn_q.weight q5_1 +blk.27.attn_v.weight q8_0 +blk.27.ffn_down_exps.weight q8_0 +blk.27.ffn_gate_exps.weight q5_1 +blk.27.ffn_up_exps.weight q5_1 +blk.28.attn_k.weight q5_1 +blk.28.attn_q.weight q5_1 +blk.28.attn_v.weight q5_1 +blk.28.ffn_down_exps.weight q5_1 +blk.28.ffn_gate_exps.weight q5_1 +blk.28.ffn_up_exps.weight q5_1 +blk.29.attn_k.weight q5_1 +blk.29.attn_q.weight q5_1 +blk.29.attn_v.weight q5_1 +blk.29.ffn_down_exps.weight q5_1 +blk.29.ffn_gate_exps.weight q5_1 +blk.29.ffn_up_exps.weight q5_1 +blk.30.attn_k.weight q5_1 +blk.30.attn_q.weight q5_1 +blk.30.attn_v.weight q8_0 +blk.30.ffn_down_exps.weight q8_0 +blk.30.ffn_gate_exps.weight q5_1 +blk.30.ffn_up_exps.weight q5_1 +blk.31.attn_k.weight q5_1 +blk.31.attn_q.weight q5_1 +blk.31.attn_v.weight q8_0 +blk.31.ffn_down_exps.weight q8_0 +blk.31.ffn_gate_exps.weight q5_1 +blk.31.ffn_up_exps.weight q5_1 +blk.32.attn_k.weight q5_1 +blk.32.attn_q.weight q5_1 +blk.32.attn_v.weight q8_0 +blk.32.ffn_down_exps.weight q8_0 +blk.32.ffn_gate_exps.weight q5_1 +blk.32.ffn_up_exps.weight q5_1 +blk.33.attn_k.weight q5_1 +blk.33.attn_q.weight q5_1 +blk.33.attn_v.weight q8_0 +blk.33.ffn_down_exps.weight q8_0 +blk.33.ffn_gate_exps.weight q5_1 +blk.33.ffn_up_exps.weight q5_1 +blk.34.attn_k.weight q5_1 +blk.34.attn_q.weight q5_1 +blk.34.attn_v.weight q8_0 +blk.34.ffn_down_exps.weight q8_0 +blk.34.ffn_gate_exps.weight q5_1 +blk.34.ffn_up_exps.weight q5_1 +blk.35.attn_k.weight q5_1 +blk.35.attn_q.weight q5_1 +blk.35.attn_v.weight q8_0 +blk.35.ffn_down_exps.weight q8_0 +blk.35.ffn_gate_exps.weight q5_1 +blk.35.ffn_up_exps.weight q5_1 [Q6_K] q6_K output.weight q8_0 +token_embd.weight q8_0 +blk.0.attn_k.weight q8_0 +blk.0.attn_q.weight q8_0 +blk.0.attn_v.weight q8_0 +blk.0.ffn_down_exps.weight q8_0 +blk.0.ffn_gate_exps.weight q8_0 +blk.0.ffn_up_exps.weight q8_0 +blk.1.attn_k.weight q8_0 +blk.1.attn_q.weight q8_0 +blk.1.attn_v.weight q8_0 +blk.1.ffn_down_exps.weight q8_0 +blk.1.ffn_gate_exps.weight q8_0 +blk.1.ffn_up_exps.weight q8_0 +blk.2.attn_k.weight q8_0 +blk.2.attn_q.weight q8_0 +blk.2.attn_v.weight q8_0 +blk.2.ffn_down_exps.weight q8_0 +blk.2.ffn_gate_exps.weight q8_0 +blk.2.ffn_up_exps.weight q8_0 +blk.3.attn_k.weight q8_0 +blk.3.attn_q.weight q8_0 +blk.3.attn_v.weight q8_0 +blk.3.ffn_down_exps.weight q8_0 +blk.3.ffn_gate_exps.weight q8_0 +blk.3.ffn_up_exps.weight q8_0 +blk.4.attn_k.weight q8_0 +blk.4.attn_q.weight q8_0 +blk.4.attn_v.weight q8_0 +blk.4.ffn_down_exps.weight q8_0 +blk.4.ffn_gate_exps.weight q8_0 +blk.4.ffn_up_exps.weight q8_0 +blk.5.attn_k.weight q8_0 +blk.5.attn_q.weight q8_0 +blk.5.attn_v.weight q8_0 +blk.5.ffn_down_exps.weight q8_0 +blk.5.ffn_gate_exps.weight q8_0 +blk.5.ffn_up_exps.weight q8_0 +blk.6.attn_k.weight q8_0 +blk.6.attn_q.weight q8_0 +blk.6.attn_v.weight q8_0 +blk.6.ffn_down_exps.weight q8_0 +blk.6.ffn_gate_exps.weight q8_0 +blk.6.ffn_up_exps.weight q8_0 +blk.7.attn_k.weight q8_0 +blk.7.attn_q.weight q8_0 +blk.7.attn_v.weight q8_0 +blk.7.ffn_down_exps.weight q8_0 +blk.7.ffn_gate_exps.weight q8_0 +blk.7.ffn_up_exps.weight q8_0 +blk.8.attn_k.weight q8_0 +blk.8.attn_q.weight q8_0 +blk.8.attn_v.weight q8_0 +blk.8.ffn_down_exps.weight q8_0 +blk.8.ffn_gate_exps.weight q8_0 +blk.8.ffn_up_exps.weight q8_0 +blk.9.attn_k.weight q8_0 +blk.9.attn_q.weight q8_0 +blk.9.attn_v.weight q8_0 +blk.9.ffn_down_exps.weight q8_0 +blk.9.ffn_gate_exps.weight q8_0 +blk.9.ffn_up_exps.weight q8_0 +blk.10.attn_k.weight q8_0 +blk.10.attn_q.weight q8_0 +blk.10.attn_v.weight q8_0 +blk.10.ffn_down_exps.weight q8_0 +blk.10.ffn_gate_exps.weight q8_0 +blk.10.ffn_up_exps.weight q8_0 +blk.11.attn_k.weight q8_0 +blk.11.attn_q.weight q8_0 +blk.11.attn_v.weight q8_0 +blk.11.ffn_down_exps.weight q8_0 +blk.11.ffn_gate_exps.weight q8_0 +blk.11.ffn_up_exps.weight q8_0 +blk.12.attn_k.weight q8_0 +blk.12.attn_q.weight q8_0 +blk.12.attn_v.weight q8_0 +blk.12.ffn_down_exps.weight q8_0 +blk.12.ffn_gate_exps.weight q8_0 +blk.12.ffn_up_exps.weight q8_0 +blk.13.attn_k.weight q8_0 +blk.13.attn_q.weight q8_0 +blk.13.attn_v.weight q8_0 +blk.13.ffn_down_exps.weight q8_0 +blk.13.ffn_gate_exps.weight q8_0 +blk.13.ffn_up_exps.weight q8_0 +blk.14.attn_k.weight q8_0 +blk.14.attn_q.weight q8_0 +blk.14.attn_v.weight q8_0 +blk.14.ffn_down_exps.weight q8_0 +blk.14.ffn_gate_exps.weight q8_0 +blk.14.ffn_up_exps.weight q8_0 +blk.15.attn_k.weight q8_0 +blk.15.attn_q.weight q8_0 +blk.15.attn_v.weight q8_0 +blk.15.ffn_down_exps.weight q8_0 +blk.15.ffn_gate_exps.weight q8_0 +blk.15.ffn_up_exps.weight q8_0 +blk.16.attn_k.weight q8_0 +blk.16.attn_q.weight q8_0 +blk.16.attn_v.weight q8_0 +blk.16.ffn_down_exps.weight q8_0 +blk.16.ffn_gate_exps.weight q8_0 +blk.16.ffn_up_exps.weight q8_0 +blk.17.attn_k.weight q8_0 +blk.17.attn_q.weight q8_0 +blk.17.attn_v.weight q8_0 +blk.17.ffn_down_exps.weight q8_0 +blk.17.ffn_gate_exps.weight q8_0 +blk.17.ffn_up_exps.weight q8_0 +blk.18.attn_k.weight q8_0 +blk.18.attn_q.weight q8_0 +blk.18.attn_v.weight q8_0 +blk.18.ffn_down_exps.weight q8_0 +blk.18.ffn_gate_exps.weight q8_0 +blk.18.ffn_up_exps.weight q8_0 +blk.19.attn_k.weight q8_0 +blk.19.attn_q.weight q8_0 +blk.19.attn_v.weight q8_0 +blk.19.ffn_down_exps.weight q8_0 +blk.19.ffn_gate_exps.weight q8_0 +blk.19.ffn_up_exps.weight q8_0 +blk.20.attn_k.weight q8_0 +blk.20.attn_q.weight q8_0 +blk.20.attn_v.weight q8_0 +blk.20.ffn_down_exps.weight q8_0 +blk.20.ffn_gate_exps.weight q8_0 +blk.20.ffn_up_exps.weight q8_0 +blk.21.attn_k.weight q8_0 +blk.21.attn_q.weight q8_0 +blk.21.attn_v.weight q8_0 +blk.21.ffn_down_exps.weight q8_0 +blk.21.ffn_gate_exps.weight q8_0 +blk.21.ffn_up_exps.weight q8_0 +blk.22.attn_k.weight q8_0 +blk.22.attn_q.weight q8_0 +blk.22.attn_v.weight q8_0 +blk.22.ffn_down_exps.weight q8_0 +blk.22.ffn_gate_exps.weight q8_0 +blk.22.ffn_up_exps.weight q8_0 +blk.23.attn_k.weight q8_0 +blk.23.attn_q.weight q8_0 +blk.23.attn_v.weight q8_0 +blk.23.ffn_down_exps.weight q8_0 +blk.23.ffn_gate_exps.weight q8_0 +blk.23.ffn_up_exps.weight q8_0 +blk.24.attn_k.weight q8_0 +blk.24.attn_q.weight q8_0 +blk.24.attn_v.weight q8_0 +blk.24.ffn_down_exps.weight q8_0 +blk.24.ffn_gate_exps.weight q8_0 +blk.24.ffn_up_exps.weight q8_0 +blk.25.attn_k.weight q8_0 +blk.25.attn_q.weight q8_0 +blk.25.attn_v.weight q8_0 +blk.25.ffn_down_exps.weight q8_0 +blk.25.ffn_gate_exps.weight q8_0 +blk.25.ffn_up_exps.weight q8_0 +blk.26.attn_k.weight q8_0 +blk.26.attn_q.weight q8_0 +blk.26.attn_v.weight q8_0 +blk.26.ffn_down_exps.weight q8_0 +blk.26.ffn_gate_exps.weight q8_0 +blk.26.ffn_up_exps.weight q8_0 +blk.27.attn_k.weight q8_0 +blk.27.attn_q.weight q8_0 +blk.27.attn_v.weight q8_0 +blk.27.ffn_down_exps.weight q8_0 +blk.27.ffn_gate_exps.weight q8_0 +blk.27.ffn_up_exps.weight q8_0 +blk.28.attn_k.weight q8_0 +blk.28.attn_q.weight q8_0 +blk.28.attn_v.weight q8_0 +blk.28.ffn_down_exps.weight q8_0 +blk.28.ffn_gate_exps.weight q8_0 +blk.28.ffn_up_exps.weight q8_0 +blk.29.attn_k.weight q8_0 +blk.29.attn_q.weight q8_0 +blk.29.attn_v.weight q8_0 +blk.29.ffn_down_exps.weight q8_0 +blk.29.ffn_gate_exps.weight q8_0 +blk.29.ffn_up_exps.weight q8_0 +blk.30.attn_k.weight q8_0 +blk.30.attn_q.weight q8_0 +blk.30.attn_v.weight q8_0 +blk.30.ffn_down_exps.weight q8_0 +blk.30.ffn_gate_exps.weight q8_0 +blk.30.ffn_up_exps.weight q8_0 +blk.31.attn_k.weight q8_0 +blk.31.attn_q.weight q8_0 +blk.31.attn_v.weight q8_0 +blk.31.ffn_down_exps.weight q8_0 +blk.31.ffn_gate_exps.weight q8_0 +blk.31.ffn_up_exps.weight q8_0 +blk.32.attn_k.weight q8_0 +blk.32.attn_q.weight q8_0 +blk.32.attn_v.weight q8_0 +blk.32.ffn_down_exps.weight q8_0 +blk.32.ffn_gate_exps.weight q8_0 +blk.32.ffn_up_exps.weight q8_0 +blk.33.attn_k.weight q8_0 +blk.33.attn_q.weight q8_0 +blk.33.attn_v.weight q8_0 +blk.33.ffn_down_exps.weight q8_0 +blk.33.ffn_gate_exps.weight q8_0 +blk.33.ffn_up_exps.weight q8_0 +blk.34.attn_k.weight q8_0 +blk.34.attn_q.weight q8_0 +blk.34.attn_v.weight q8_0 +blk.34.ffn_down_exps.weight q8_0 +blk.34.ffn_gate_exps.weight q8_0 +blk.34.ffn_up_exps.weight q8_0 +blk.35.attn_k.weight q8_0 +blk.35.attn_q.weight q8_0 +blk.35.attn_v.weight q8_0 +blk.35.ffn_down_exps.weight q8_0 +blk.35.ffn_gate_exps.weight q8_0 +blk.35.ffn_up_exps.weight q8_0 [IQ2_XXS] iq2_xxs output.weight q8_0 -token_embd.weight q2_K -blk.0.attn_v.weight q4_K -blk.0.ffn_down_exps.weight q2_K -blk.1.attn_v.weight q4_K -blk.1.ffn_down_exps.weight q2_K -blk.2.attn_v.weight q4_K -blk.2.ffn_down_exps.weight q2_K -blk.3.attn_v.weight q4_K -blk.3.ffn_down_exps.weight q2_K -blk.4.attn_v.weight q4_K -blk.5.attn_v.weight q4_K -blk.6.attn_v.weight q4_K -blk.7.attn_v.weight q4_K -blk.8.attn_v.weight q4_K -blk.9.attn_v.weight q4_K -blk.10.attn_v.weight q4_K -blk.11.attn_v.weight q4_K -blk.12.attn_v.weight q4_K -blk.13.attn_v.weight q4_K -blk.14.attn_v.weight q4_K -blk.15.attn_v.weight q4_K -blk.16.attn_v.weight q4_K -blk.17.attn_v.weight q4_K -blk.18.attn_v.weight q4_K -blk.19.attn_v.weight q4_K -blk.20.attn_v.weight q4_K -blk.21.attn_v.weight q4_K -blk.22.attn_v.weight q4_K -blk.23.attn_v.weight q4_K -blk.24.attn_v.weight q4_K -blk.25.attn_v.weight q4_K -blk.26.attn_v.weight q4_K -blk.27.attn_v.weight q4_K -blk.28.attn_v.weight q4_K -blk.29.attn_v.weight q4_K -blk.30.attn_v.weight q4_K -blk.31.attn_v.weight q4_K -blk.32.attn_v.weight q4_K -blk.33.attn_v.weight q4_K -blk.34.attn_v.weight q4_K -blk.35.attn_v.weight q4_K +token_embd.weight q4_0 +blk.0.attn_k.weight iq4_nl +blk.0.attn_q.weight iq4_nl +blk.0.attn_v.weight q5_0 +blk.0.ffn_down_exps.weight q4_0 +blk.0.ffn_gate_exps.weight iq4_nl +blk.0.ffn_up_exps.weight iq4_nl +blk.1.attn_k.weight iq4_nl +blk.1.attn_q.weight iq4_nl +blk.1.attn_v.weight q5_0 +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_gate_exps.weight iq4_nl +blk.1.ffn_up_exps.weight iq4_nl +blk.2.attn_k.weight iq4_nl +blk.2.attn_q.weight iq4_nl +blk.2.attn_v.weight q5_0 +blk.2.ffn_down_exps.weight q4_0 +blk.2.ffn_gate_exps.weight iq4_nl +blk.2.ffn_up_exps.weight iq4_nl +blk.3.attn_k.weight iq4_nl +blk.3.attn_q.weight iq4_nl +blk.3.attn_v.weight q5_0 +blk.3.ffn_down_exps.weight q4_0 +blk.3.ffn_gate_exps.weight iq4_nl +blk.3.ffn_up_exps.weight iq4_nl +blk.4.attn_k.weight iq4_nl +blk.4.attn_q.weight iq4_nl +blk.4.attn_v.weight q5_0 +blk.4.ffn_down_exps.weight iq4_nl +blk.4.ffn_gate_exps.weight iq4_nl +blk.4.ffn_up_exps.weight iq4_nl +blk.5.attn_k.weight iq4_nl +blk.5.attn_q.weight iq4_nl +blk.5.attn_v.weight q5_0 +blk.5.ffn_down_exps.weight iq4_nl +blk.5.ffn_gate_exps.weight iq4_nl +blk.5.ffn_up_exps.weight iq4_nl +blk.6.attn_k.weight iq4_nl +blk.6.attn_q.weight iq4_nl +blk.6.attn_v.weight q5_0 +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_gate_exps.weight iq4_nl +blk.6.ffn_up_exps.weight iq4_nl +blk.7.attn_k.weight iq4_nl +blk.7.attn_q.weight iq4_nl +blk.7.attn_v.weight q5_0 +blk.7.ffn_down_exps.weight iq4_nl +blk.7.ffn_gate_exps.weight iq4_nl +blk.7.ffn_up_exps.weight iq4_nl +blk.8.attn_k.weight iq4_nl +blk.8.attn_q.weight iq4_nl +blk.8.attn_v.weight q5_0 +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_gate_exps.weight iq4_nl +blk.8.ffn_up_exps.weight iq4_nl +blk.9.attn_k.weight iq4_nl +blk.9.attn_q.weight iq4_nl +blk.9.attn_v.weight q5_0 +blk.9.ffn_down_exps.weight iq4_nl +blk.9.ffn_gate_exps.weight iq4_nl +blk.9.ffn_up_exps.weight iq4_nl +blk.10.attn_k.weight iq4_nl +blk.10.attn_q.weight iq4_nl +blk.10.attn_v.weight q5_0 +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_gate_exps.weight iq4_nl +blk.10.ffn_up_exps.weight iq4_nl +blk.11.attn_k.weight iq4_nl +blk.11.attn_q.weight iq4_nl +blk.11.attn_v.weight q5_0 +blk.11.ffn_down_exps.weight iq4_nl +blk.11.ffn_gate_exps.weight iq4_nl +blk.11.ffn_up_exps.weight iq4_nl +blk.12.attn_k.weight iq4_nl +blk.12.attn_q.weight iq4_nl +blk.12.attn_v.weight q5_0 +blk.12.ffn_down_exps.weight iq4_nl +blk.12.ffn_gate_exps.weight iq4_nl +blk.12.ffn_up_exps.weight iq4_nl +blk.13.attn_k.weight iq4_nl +blk.13.attn_q.weight iq4_nl +blk.13.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_gate_exps.weight iq4_nl +blk.13.ffn_up_exps.weight iq4_nl +blk.14.attn_k.weight iq4_nl +blk.14.attn_q.weight iq4_nl +blk.14.attn_v.weight q5_0 +blk.14.ffn_down_exps.weight iq4_nl +blk.14.ffn_gate_exps.weight iq4_nl +blk.14.ffn_up_exps.weight iq4_nl +blk.15.attn_k.weight iq4_nl +blk.15.attn_q.weight iq4_nl +blk.15.attn_v.weight q5_0 +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_gate_exps.weight iq4_nl +blk.15.ffn_up_exps.weight iq4_nl +blk.16.attn_k.weight iq4_nl +blk.16.attn_q.weight iq4_nl +blk.16.attn_v.weight q5_0 +blk.16.ffn_down_exps.weight iq4_nl +blk.16.ffn_gate_exps.weight iq4_nl +blk.16.ffn_up_exps.weight iq4_nl +blk.17.attn_k.weight iq4_nl +blk.17.attn_q.weight iq4_nl +blk.17.attn_v.weight q5_0 +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_gate_exps.weight iq4_nl +blk.17.ffn_up_exps.weight iq4_nl +blk.18.attn_k.weight iq4_nl +blk.18.attn_q.weight iq4_nl +blk.18.attn_v.weight q5_0 +blk.18.ffn_down_exps.weight iq4_nl +blk.18.ffn_gate_exps.weight iq4_nl +blk.18.ffn_up_exps.weight iq4_nl +blk.19.attn_k.weight iq4_nl +blk.19.attn_q.weight iq4_nl +blk.19.attn_v.weight q5_0 +blk.19.ffn_down_exps.weight iq4_nl +blk.19.ffn_gate_exps.weight iq4_nl +blk.19.ffn_up_exps.weight iq4_nl +blk.20.attn_k.weight iq4_nl +blk.20.attn_q.weight iq4_nl +blk.20.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_gate_exps.weight iq4_nl +blk.20.ffn_up_exps.weight iq4_nl +blk.21.attn_k.weight iq4_nl +blk.21.attn_q.weight iq4_nl +blk.21.attn_v.weight q5_0 +blk.21.ffn_down_exps.weight iq4_nl +blk.21.ffn_gate_exps.weight iq4_nl +blk.21.ffn_up_exps.weight iq4_nl +blk.22.attn_k.weight iq4_nl +blk.22.attn_q.weight iq4_nl +blk.22.attn_v.weight q5_0 +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_gate_exps.weight iq4_nl +blk.22.ffn_up_exps.weight iq4_nl +blk.23.attn_k.weight iq4_nl +blk.23.attn_q.weight iq4_nl +blk.23.attn_v.weight q5_0 +blk.23.ffn_down_exps.weight iq4_nl +blk.23.ffn_gate_exps.weight iq4_nl +blk.23.ffn_up_exps.weight iq4_nl +blk.24.attn_k.weight iq4_nl +blk.24.attn_q.weight iq4_nl +blk.24.attn_v.weight q5_0 +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_gate_exps.weight iq4_nl +blk.24.ffn_up_exps.weight iq4_nl +blk.25.attn_k.weight iq4_nl +blk.25.attn_q.weight iq4_nl +blk.25.attn_v.weight q5_0 +blk.25.ffn_down_exps.weight iq4_nl +blk.25.ffn_gate_exps.weight iq4_nl +blk.25.ffn_up_exps.weight iq4_nl +blk.26.attn_k.weight iq4_nl +blk.26.attn_q.weight iq4_nl +blk.26.attn_v.weight q5_0 +blk.26.ffn_down_exps.weight iq4_nl +blk.26.ffn_gate_exps.weight iq4_nl +blk.26.ffn_up_exps.weight iq4_nl +blk.27.attn_k.weight iq4_nl +blk.27.attn_q.weight iq4_nl +blk.27.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_gate_exps.weight iq4_nl +blk.27.ffn_up_exps.weight iq4_nl +blk.28.attn_k.weight iq4_nl +blk.28.attn_q.weight iq4_nl +blk.28.attn_v.weight q5_0 +blk.28.ffn_down_exps.weight iq4_nl +blk.28.ffn_gate_exps.weight iq4_nl +blk.28.ffn_up_exps.weight iq4_nl +blk.29.attn_k.weight iq4_nl +blk.29.attn_q.weight iq4_nl +blk.29.attn_v.weight q5_0 +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_gate_exps.weight iq4_nl +blk.29.ffn_up_exps.weight iq4_nl +blk.30.attn_k.weight iq4_nl +blk.30.attn_q.weight iq4_nl +blk.30.attn_v.weight q5_0 +blk.30.ffn_down_exps.weight iq4_nl +blk.30.ffn_gate_exps.weight iq4_nl +blk.30.ffn_up_exps.weight iq4_nl +blk.31.attn_k.weight iq4_nl +blk.31.attn_q.weight iq4_nl +blk.31.attn_v.weight q5_0 +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_gate_exps.weight iq4_nl +blk.31.ffn_up_exps.weight iq4_nl +blk.32.attn_k.weight iq4_nl +blk.32.attn_q.weight iq4_nl +blk.32.attn_v.weight q5_0 +blk.32.ffn_down_exps.weight iq4_nl +blk.32.ffn_gate_exps.weight iq4_nl +blk.32.ffn_up_exps.weight iq4_nl +blk.33.attn_k.weight iq4_nl +blk.33.attn_q.weight iq4_nl +blk.33.attn_v.weight q5_0 +blk.33.ffn_down_exps.weight iq4_nl +blk.33.ffn_gate_exps.weight iq4_nl +blk.33.ffn_up_exps.weight iq4_nl +blk.34.attn_k.weight iq4_nl +blk.34.attn_q.weight iq4_nl +blk.34.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_gate_exps.weight iq4_nl +blk.34.ffn_up_exps.weight iq4_nl +blk.35.attn_k.weight iq4_nl +blk.35.attn_q.weight iq4_nl +blk.35.attn_v.weight q5_0 +blk.35.ffn_down_exps.weight iq4_nl +blk.35.ffn_gate_exps.weight iq4_nl +blk.35.ffn_up_exps.weight iq4_nl [IQ2_XS] iq2_xs output.weight q8_0 -token_embd.weight q2_K -blk.0.attn_v.weight q4_K -blk.0.ffn_down_exps.weight q2_K -blk.1.attn_v.weight q4_K -blk.1.ffn_down_exps.weight q2_K -blk.2.attn_v.weight q4_K -blk.2.ffn_down_exps.weight q2_K -blk.3.attn_v.weight q4_K -blk.3.ffn_down_exps.weight q2_K -blk.4.attn_v.weight q4_K -blk.5.attn_v.weight q4_K -blk.6.attn_v.weight q4_K -blk.7.attn_v.weight q4_K -blk.8.attn_v.weight q4_K -blk.9.attn_v.weight q4_K -blk.10.attn_v.weight q4_K -blk.11.attn_v.weight q4_K -blk.12.attn_v.weight q4_K -blk.13.attn_v.weight q4_K -blk.14.attn_v.weight q4_K -blk.15.attn_v.weight q4_K -blk.16.attn_v.weight q4_K -blk.17.attn_v.weight q4_K -blk.18.attn_v.weight q4_K -blk.19.attn_v.weight q4_K -blk.20.attn_v.weight q4_K -blk.21.attn_v.weight q4_K -blk.22.attn_v.weight q4_K -blk.23.attn_v.weight q4_K -blk.24.attn_v.weight q4_K -blk.25.attn_v.weight q4_K -blk.26.attn_v.weight q4_K -blk.27.attn_v.weight q4_K -blk.28.attn_v.weight q4_K -blk.29.attn_v.weight q4_K -blk.30.attn_v.weight q4_K -blk.31.attn_v.weight q4_K -blk.32.attn_v.weight q4_K -blk.33.attn_v.weight q4_K -blk.34.attn_v.weight q4_K -blk.35.attn_v.weight q4_K +token_embd.weight q4_0 +blk.0.attn_k.weight iq4_nl +blk.0.attn_q.weight iq4_nl +blk.0.attn_v.weight q5_0 +blk.0.ffn_down_exps.weight q4_0 +blk.0.ffn_gate_exps.weight iq4_nl +blk.0.ffn_up_exps.weight iq4_nl +blk.1.attn_k.weight iq4_nl +blk.1.attn_q.weight iq4_nl +blk.1.attn_v.weight q5_0 +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_gate_exps.weight iq4_nl +blk.1.ffn_up_exps.weight iq4_nl +blk.2.attn_k.weight iq4_nl +blk.2.attn_q.weight iq4_nl +blk.2.attn_v.weight q5_0 +blk.2.ffn_down_exps.weight q4_0 +blk.2.ffn_gate_exps.weight iq4_nl +blk.2.ffn_up_exps.weight iq4_nl +blk.3.attn_k.weight iq4_nl +blk.3.attn_q.weight iq4_nl +blk.3.attn_v.weight q5_0 +blk.3.ffn_down_exps.weight q4_0 +blk.3.ffn_gate_exps.weight iq4_nl +blk.3.ffn_up_exps.weight iq4_nl +blk.4.attn_k.weight iq4_nl +blk.4.attn_q.weight iq4_nl +blk.4.attn_v.weight q5_0 +blk.4.ffn_down_exps.weight iq4_nl +blk.4.ffn_gate_exps.weight iq4_nl +blk.4.ffn_up_exps.weight iq4_nl +blk.5.attn_k.weight iq4_nl +blk.5.attn_q.weight iq4_nl +blk.5.attn_v.weight q5_0 +blk.5.ffn_down_exps.weight iq4_nl +blk.5.ffn_gate_exps.weight iq4_nl +blk.5.ffn_up_exps.weight iq4_nl +blk.6.attn_k.weight iq4_nl +blk.6.attn_q.weight iq4_nl +blk.6.attn_v.weight q5_0 +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_gate_exps.weight iq4_nl +blk.6.ffn_up_exps.weight iq4_nl +blk.7.attn_k.weight iq4_nl +blk.7.attn_q.weight iq4_nl +blk.7.attn_v.weight q5_0 +blk.7.ffn_down_exps.weight iq4_nl +blk.7.ffn_gate_exps.weight iq4_nl +blk.7.ffn_up_exps.weight iq4_nl +blk.8.attn_k.weight iq4_nl +blk.8.attn_q.weight iq4_nl +blk.8.attn_v.weight q5_0 +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_gate_exps.weight iq4_nl +blk.8.ffn_up_exps.weight iq4_nl +blk.9.attn_k.weight iq4_nl +blk.9.attn_q.weight iq4_nl +blk.9.attn_v.weight q5_0 +blk.9.ffn_down_exps.weight iq4_nl +blk.9.ffn_gate_exps.weight iq4_nl +blk.9.ffn_up_exps.weight iq4_nl +blk.10.attn_k.weight iq4_nl +blk.10.attn_q.weight iq4_nl +blk.10.attn_v.weight q5_0 +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_gate_exps.weight iq4_nl +blk.10.ffn_up_exps.weight iq4_nl +blk.11.attn_k.weight iq4_nl +blk.11.attn_q.weight iq4_nl +blk.11.attn_v.weight q5_0 +blk.11.ffn_down_exps.weight iq4_nl +blk.11.ffn_gate_exps.weight iq4_nl +blk.11.ffn_up_exps.weight iq4_nl +blk.12.attn_k.weight iq4_nl +blk.12.attn_q.weight iq4_nl +blk.12.attn_v.weight q5_0 +blk.12.ffn_down_exps.weight iq4_nl +blk.12.ffn_gate_exps.weight iq4_nl +blk.12.ffn_up_exps.weight iq4_nl +blk.13.attn_k.weight iq4_nl +blk.13.attn_q.weight iq4_nl +blk.13.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_gate_exps.weight iq4_nl +blk.13.ffn_up_exps.weight iq4_nl +blk.14.attn_k.weight iq4_nl +blk.14.attn_q.weight iq4_nl +blk.14.attn_v.weight q5_0 +blk.14.ffn_down_exps.weight iq4_nl +blk.14.ffn_gate_exps.weight iq4_nl +blk.14.ffn_up_exps.weight iq4_nl +blk.15.attn_k.weight iq4_nl +blk.15.attn_q.weight iq4_nl +blk.15.attn_v.weight q5_0 +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_gate_exps.weight iq4_nl +blk.15.ffn_up_exps.weight iq4_nl +blk.16.attn_k.weight iq4_nl +blk.16.attn_q.weight iq4_nl +blk.16.attn_v.weight q5_0 +blk.16.ffn_down_exps.weight iq4_nl +blk.16.ffn_gate_exps.weight iq4_nl +blk.16.ffn_up_exps.weight iq4_nl +blk.17.attn_k.weight iq4_nl +blk.17.attn_q.weight iq4_nl +blk.17.attn_v.weight q5_0 +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_gate_exps.weight iq4_nl +blk.17.ffn_up_exps.weight iq4_nl +blk.18.attn_k.weight iq4_nl +blk.18.attn_q.weight iq4_nl +blk.18.attn_v.weight q5_0 +blk.18.ffn_down_exps.weight iq4_nl +blk.18.ffn_gate_exps.weight iq4_nl +blk.18.ffn_up_exps.weight iq4_nl +blk.19.attn_k.weight iq4_nl +blk.19.attn_q.weight iq4_nl +blk.19.attn_v.weight q5_0 +blk.19.ffn_down_exps.weight iq4_nl +blk.19.ffn_gate_exps.weight iq4_nl +blk.19.ffn_up_exps.weight iq4_nl +blk.20.attn_k.weight iq4_nl +blk.20.attn_q.weight iq4_nl +blk.20.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_gate_exps.weight iq4_nl +blk.20.ffn_up_exps.weight iq4_nl +blk.21.attn_k.weight iq4_nl +blk.21.attn_q.weight iq4_nl +blk.21.attn_v.weight q5_0 +blk.21.ffn_down_exps.weight iq4_nl +blk.21.ffn_gate_exps.weight iq4_nl +blk.21.ffn_up_exps.weight iq4_nl +blk.22.attn_k.weight iq4_nl +blk.22.attn_q.weight iq4_nl +blk.22.attn_v.weight q5_0 +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_gate_exps.weight iq4_nl +blk.22.ffn_up_exps.weight iq4_nl +blk.23.attn_k.weight iq4_nl +blk.23.attn_q.weight iq4_nl +blk.23.attn_v.weight q5_0 +blk.23.ffn_down_exps.weight iq4_nl +blk.23.ffn_gate_exps.weight iq4_nl +blk.23.ffn_up_exps.weight iq4_nl +blk.24.attn_k.weight iq4_nl +blk.24.attn_q.weight iq4_nl +blk.24.attn_v.weight q5_0 +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_gate_exps.weight iq4_nl +blk.24.ffn_up_exps.weight iq4_nl +blk.25.attn_k.weight iq4_nl +blk.25.attn_q.weight iq4_nl +blk.25.attn_v.weight q5_0 +blk.25.ffn_down_exps.weight iq4_nl +blk.25.ffn_gate_exps.weight iq4_nl +blk.25.ffn_up_exps.weight iq4_nl +blk.26.attn_k.weight iq4_nl +blk.26.attn_q.weight iq4_nl +blk.26.attn_v.weight q5_0 +blk.26.ffn_down_exps.weight iq4_nl +blk.26.ffn_gate_exps.weight iq4_nl +blk.26.ffn_up_exps.weight iq4_nl +blk.27.attn_k.weight iq4_nl +blk.27.attn_q.weight iq4_nl +blk.27.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_gate_exps.weight iq4_nl +blk.27.ffn_up_exps.weight iq4_nl +blk.28.attn_k.weight iq4_nl +blk.28.attn_q.weight iq4_nl +blk.28.attn_v.weight q5_0 +blk.28.ffn_down_exps.weight iq4_nl +blk.28.ffn_gate_exps.weight iq4_nl +blk.28.ffn_up_exps.weight iq4_nl +blk.29.attn_k.weight iq4_nl +blk.29.attn_q.weight iq4_nl +blk.29.attn_v.weight q5_0 +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_gate_exps.weight iq4_nl +blk.29.ffn_up_exps.weight iq4_nl +blk.30.attn_k.weight iq4_nl +blk.30.attn_q.weight iq4_nl +blk.30.attn_v.weight q5_0 +blk.30.ffn_down_exps.weight iq4_nl +blk.30.ffn_gate_exps.weight iq4_nl +blk.30.ffn_up_exps.weight iq4_nl +blk.31.attn_k.weight iq4_nl +blk.31.attn_q.weight iq4_nl +blk.31.attn_v.weight q5_0 +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_gate_exps.weight iq4_nl +blk.31.ffn_up_exps.weight iq4_nl +blk.32.attn_k.weight iq4_nl +blk.32.attn_q.weight iq4_nl +blk.32.attn_v.weight q5_0 +blk.32.ffn_down_exps.weight iq4_nl +blk.32.ffn_gate_exps.weight iq4_nl +blk.32.ffn_up_exps.weight iq4_nl +blk.33.attn_k.weight iq4_nl +blk.33.attn_q.weight iq4_nl +blk.33.attn_v.weight q5_0 +blk.33.ffn_down_exps.weight iq4_nl +blk.33.ffn_gate_exps.weight iq4_nl +blk.33.ffn_up_exps.weight iq4_nl +blk.34.attn_k.weight iq4_nl +blk.34.attn_q.weight iq4_nl +blk.34.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_gate_exps.weight iq4_nl +blk.34.ffn_up_exps.weight iq4_nl +blk.35.attn_k.weight iq4_nl +blk.35.attn_q.weight iq4_nl +blk.35.attn_v.weight q5_0 +blk.35.ffn_down_exps.weight iq4_nl +blk.35.ffn_gate_exps.weight iq4_nl +blk.35.ffn_up_exps.weight iq4_nl [Q2_K_S] q2_K output.weight q8_0 -blk.0.attn_v.weight q4_K -blk.0.ffn_down_exps.weight q4_K -blk.1.attn_v.weight q4_K -blk.1.ffn_down_exps.weight q4_K -blk.2.attn_v.weight q4_K -blk.2.ffn_down_exps.weight q4_K -blk.3.attn_v.weight q4_K -blk.3.ffn_down_exps.weight q4_K -blk.4.attn_v.weight q4_K -blk.5.attn_v.weight q4_K -blk.6.attn_v.weight q4_K -blk.7.attn_v.weight q4_K -blk.8.attn_v.weight q4_K -blk.9.attn_v.weight q4_K -blk.10.attn_v.weight q4_K -blk.11.attn_v.weight q4_K -blk.12.attn_v.weight q4_K -blk.13.attn_v.weight q4_K -blk.14.attn_v.weight q4_K -blk.15.attn_v.weight q4_K -blk.16.attn_v.weight q4_K -blk.17.attn_v.weight q4_K -blk.18.attn_v.weight q4_K -blk.19.attn_v.weight q4_K -blk.20.attn_v.weight q4_K -blk.21.attn_v.weight q4_K -blk.22.attn_v.weight q4_K -blk.23.attn_v.weight q4_K -blk.24.attn_v.weight q4_K -blk.25.attn_v.weight q4_K -blk.26.attn_v.weight q4_K -blk.27.attn_v.weight q4_K -blk.28.attn_v.weight q4_K -blk.29.attn_v.weight q4_K -blk.30.attn_v.weight q4_K -blk.31.attn_v.weight q4_K -blk.32.attn_v.weight q4_K -blk.33.attn_v.weight q4_K -blk.34.attn_v.weight q4_K -blk.35.attn_v.weight q4_K +token_embd.weight q4_0 +blk.0.attn_k.weight q4_0 +blk.0.attn_q.weight q4_0 +blk.0.attn_v.weight q5_0 +blk.0.ffn_down_exps.weight q5_0 +blk.0.ffn_gate_exps.weight q4_0 +blk.0.ffn_up_exps.weight q4_0 +blk.1.attn_k.weight q4_0 +blk.1.attn_q.weight q4_0 +blk.1.attn_v.weight q5_0 +blk.1.ffn_down_exps.weight q5_0 +blk.1.ffn_gate_exps.weight q4_0 +blk.1.ffn_up_exps.weight q4_0 +blk.2.attn_k.weight q4_0 +blk.2.attn_q.weight q4_0 +blk.2.attn_v.weight q5_0 +blk.2.ffn_down_exps.weight q5_0 +blk.2.ffn_gate_exps.weight q4_0 +blk.2.ffn_up_exps.weight q4_0 +blk.3.attn_k.weight q4_0 +blk.3.attn_q.weight q4_0 +blk.3.attn_v.weight q5_0 +blk.3.ffn_down_exps.weight q5_0 +blk.3.ffn_gate_exps.weight q4_0 +blk.3.ffn_up_exps.weight q4_0 +blk.4.attn_k.weight q4_0 +blk.4.attn_q.weight q4_0 +blk.4.attn_v.weight q5_0 +blk.4.ffn_down_exps.weight q4_0 +blk.4.ffn_gate_exps.weight q4_0 +blk.4.ffn_up_exps.weight q4_0 +blk.5.attn_k.weight q4_0 +blk.5.attn_q.weight q4_0 +blk.5.attn_v.weight q5_0 +blk.5.ffn_down_exps.weight q4_0 +blk.5.ffn_gate_exps.weight q4_0 +blk.5.ffn_up_exps.weight q4_0 +blk.6.attn_k.weight q4_0 +blk.6.attn_q.weight q4_0 +blk.6.attn_v.weight q5_0 +blk.6.ffn_down_exps.weight q4_0 +blk.6.ffn_gate_exps.weight q4_0 +blk.6.ffn_up_exps.weight q4_0 +blk.7.attn_k.weight q4_0 +blk.7.attn_q.weight q4_0 +blk.7.attn_v.weight q5_0 +blk.7.ffn_down_exps.weight q4_0 +blk.7.ffn_gate_exps.weight q4_0 +blk.7.ffn_up_exps.weight q4_0 +blk.8.attn_k.weight q4_0 +blk.8.attn_q.weight q4_0 +blk.8.attn_v.weight q5_0 +blk.8.ffn_down_exps.weight q4_0 +blk.8.ffn_gate_exps.weight q4_0 +blk.8.ffn_up_exps.weight q4_0 +blk.9.attn_k.weight q4_0 +blk.9.attn_q.weight q4_0 +blk.9.attn_v.weight q5_0 +blk.9.ffn_down_exps.weight q4_0 +blk.9.ffn_gate_exps.weight q4_0 +blk.9.ffn_up_exps.weight q4_0 +blk.10.attn_k.weight q4_0 +blk.10.attn_q.weight q4_0 +blk.10.attn_v.weight q5_0 +blk.10.ffn_down_exps.weight q4_0 +blk.10.ffn_gate_exps.weight q4_0 +blk.10.ffn_up_exps.weight q4_0 +blk.11.attn_k.weight q4_0 +blk.11.attn_q.weight q4_0 +blk.11.attn_v.weight q5_0 +blk.11.ffn_down_exps.weight q4_0 +blk.11.ffn_gate_exps.weight q4_0 +blk.11.ffn_up_exps.weight q4_0 +blk.12.attn_k.weight q4_0 +blk.12.attn_q.weight q4_0 +blk.12.attn_v.weight q5_0 +blk.12.ffn_down_exps.weight q4_0 +blk.12.ffn_gate_exps.weight q4_0 +blk.12.ffn_up_exps.weight q4_0 +blk.13.attn_k.weight q4_0 +blk.13.attn_q.weight q4_0 +blk.13.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight q4_0 +blk.13.ffn_gate_exps.weight q4_0 +blk.13.ffn_up_exps.weight q4_0 +blk.14.attn_k.weight q4_0 +blk.14.attn_q.weight q4_0 +blk.14.attn_v.weight q5_0 +blk.14.ffn_down_exps.weight q4_0 +blk.14.ffn_gate_exps.weight q4_0 +blk.14.ffn_up_exps.weight q4_0 +blk.15.attn_k.weight q4_0 +blk.15.attn_q.weight q4_0 +blk.15.attn_v.weight q5_0 +blk.15.ffn_down_exps.weight q4_0 +blk.15.ffn_gate_exps.weight q4_0 +blk.15.ffn_up_exps.weight q4_0 +blk.16.attn_k.weight q4_0 +blk.16.attn_q.weight q4_0 +blk.16.attn_v.weight q5_0 +blk.16.ffn_down_exps.weight q4_0 +blk.16.ffn_gate_exps.weight q4_0 +blk.16.ffn_up_exps.weight q4_0 +blk.17.attn_k.weight q4_0 +blk.17.attn_q.weight q4_0 +blk.17.attn_v.weight q5_0 +blk.17.ffn_down_exps.weight q4_0 +blk.17.ffn_gate_exps.weight q4_0 +blk.17.ffn_up_exps.weight q4_0 +blk.18.attn_k.weight q4_0 +blk.18.attn_q.weight q4_0 +blk.18.attn_v.weight q5_0 +blk.18.ffn_down_exps.weight q4_0 +blk.18.ffn_gate_exps.weight q4_0 +blk.18.ffn_up_exps.weight q4_0 +blk.19.attn_k.weight q4_0 +blk.19.attn_q.weight q4_0 +blk.19.attn_v.weight q5_0 +blk.19.ffn_down_exps.weight q4_0 +blk.19.ffn_gate_exps.weight q4_0 +blk.19.ffn_up_exps.weight q4_0 +blk.20.attn_k.weight q4_0 +blk.20.attn_q.weight q4_0 +blk.20.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight q4_0 +blk.20.ffn_gate_exps.weight q4_0 +blk.20.ffn_up_exps.weight q4_0 +blk.21.attn_k.weight q4_0 +blk.21.attn_q.weight q4_0 +blk.21.attn_v.weight q5_0 +blk.21.ffn_down_exps.weight q4_0 +blk.21.ffn_gate_exps.weight q4_0 +blk.21.ffn_up_exps.weight q4_0 +blk.22.attn_k.weight q4_0 +blk.22.attn_q.weight q4_0 +blk.22.attn_v.weight q5_0 +blk.22.ffn_down_exps.weight q4_0 +blk.22.ffn_gate_exps.weight q4_0 +blk.22.ffn_up_exps.weight q4_0 +blk.23.attn_k.weight q4_0 +blk.23.attn_q.weight q4_0 +blk.23.attn_v.weight q5_0 +blk.23.ffn_down_exps.weight q4_0 +blk.23.ffn_gate_exps.weight q4_0 +blk.23.ffn_up_exps.weight q4_0 +blk.24.attn_k.weight q4_0 +blk.24.attn_q.weight q4_0 +blk.24.attn_v.weight q5_0 +blk.24.ffn_down_exps.weight q4_0 +blk.24.ffn_gate_exps.weight q4_0 +blk.24.ffn_up_exps.weight q4_0 +blk.25.attn_k.weight q4_0 +blk.25.attn_q.weight q4_0 +blk.25.attn_v.weight q5_0 +blk.25.ffn_down_exps.weight q4_0 +blk.25.ffn_gate_exps.weight q4_0 +blk.25.ffn_up_exps.weight q4_0 +blk.26.attn_k.weight q4_0 +blk.26.attn_q.weight q4_0 +blk.26.attn_v.weight q5_0 +blk.26.ffn_down_exps.weight q4_0 +blk.26.ffn_gate_exps.weight q4_0 +blk.26.ffn_up_exps.weight q4_0 +blk.27.attn_k.weight q4_0 +blk.27.attn_q.weight q4_0 +blk.27.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight q4_0 +blk.27.ffn_gate_exps.weight q4_0 +blk.27.ffn_up_exps.weight q4_0 +blk.28.attn_k.weight q4_0 +blk.28.attn_q.weight q4_0 +blk.28.attn_v.weight q5_0 +blk.28.ffn_down_exps.weight q4_0 +blk.28.ffn_gate_exps.weight q4_0 +blk.28.ffn_up_exps.weight q4_0 +blk.29.attn_k.weight q4_0 +blk.29.attn_q.weight q4_0 +blk.29.attn_v.weight q5_0 +blk.29.ffn_down_exps.weight q4_0 +blk.29.ffn_gate_exps.weight q4_0 +blk.29.ffn_up_exps.weight q4_0 +blk.30.attn_k.weight q4_0 +blk.30.attn_q.weight q4_0 +blk.30.attn_v.weight q5_0 +blk.30.ffn_down_exps.weight q4_0 +blk.30.ffn_gate_exps.weight q4_0 +blk.30.ffn_up_exps.weight q4_0 +blk.31.attn_k.weight q4_0 +blk.31.attn_q.weight q4_0 +blk.31.attn_v.weight q5_0 +blk.31.ffn_down_exps.weight q4_0 +blk.31.ffn_gate_exps.weight q4_0 +blk.31.ffn_up_exps.weight q4_0 +blk.32.attn_k.weight q4_0 +blk.32.attn_q.weight q4_0 +blk.32.attn_v.weight q5_0 +blk.32.ffn_down_exps.weight q4_0 +blk.32.ffn_gate_exps.weight q4_0 +blk.32.ffn_up_exps.weight q4_0 +blk.33.attn_k.weight q4_0 +blk.33.attn_q.weight q4_0 +blk.33.attn_v.weight q5_0 +blk.33.ffn_down_exps.weight q4_0 +blk.33.ffn_gate_exps.weight q4_0 +blk.33.ffn_up_exps.weight q4_0 +blk.34.attn_k.weight q4_0 +blk.34.attn_q.weight q4_0 +blk.34.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight q4_0 +blk.34.ffn_gate_exps.weight q4_0 +blk.34.ffn_up_exps.weight q4_0 +blk.35.attn_k.weight q4_0 +blk.35.attn_q.weight q4_0 +blk.35.attn_v.weight q5_0 +blk.35.ffn_down_exps.weight q4_0 +blk.35.ffn_gate_exps.weight q4_0 +blk.35.ffn_up_exps.weight q4_0 [IQ3_XS] iq3_s output.weight q8_0 -blk.0.attn_k.weight iq3_xxs -blk.0.attn_q.weight iq3_xxs -blk.0.attn_v.weight q4_K -blk.1.attn_k.weight iq3_xxs -blk.1.attn_q.weight iq3_xxs -blk.1.attn_v.weight q4_K -blk.2.attn_k.weight iq3_xxs -blk.2.attn_q.weight iq3_xxs -blk.2.attn_v.weight q4_K -blk.3.attn_k.weight iq3_xxs -blk.3.attn_q.weight iq3_xxs -blk.3.attn_v.weight q4_K -blk.4.attn_k.weight iq3_xxs -blk.4.attn_q.weight iq3_xxs -blk.4.attn_v.weight q4_K -blk.4.ffn_gate_exps.weight iq3_xxs -blk.4.ffn_up_exps.weight iq3_xxs -blk.5.attn_k.weight iq3_xxs -blk.5.attn_q.weight iq3_xxs -blk.5.attn_v.weight q4_K -blk.5.ffn_gate_exps.weight iq3_xxs -blk.5.ffn_up_exps.weight iq3_xxs -blk.6.attn_k.weight iq3_xxs -blk.6.attn_q.weight iq3_xxs -blk.6.attn_v.weight q4_K -blk.6.ffn_gate_exps.weight iq3_xxs -blk.6.ffn_up_exps.weight iq3_xxs -blk.7.attn_k.weight iq3_xxs -blk.7.attn_q.weight iq3_xxs -blk.7.attn_v.weight q4_K -blk.7.ffn_gate_exps.weight iq3_xxs -blk.7.ffn_up_exps.weight iq3_xxs -blk.8.attn_k.weight iq3_xxs -blk.8.attn_q.weight iq3_xxs -blk.8.attn_v.weight q4_K -blk.8.ffn_gate_exps.weight iq3_xxs -blk.8.ffn_up_exps.weight iq3_xxs -blk.9.attn_k.weight iq3_xxs -blk.9.attn_q.weight iq3_xxs -blk.9.attn_v.weight q4_K -blk.9.ffn_gate_exps.weight iq3_xxs -blk.9.ffn_up_exps.weight iq3_xxs -blk.10.attn_k.weight iq3_xxs -blk.10.attn_q.weight iq3_xxs -blk.10.attn_v.weight q4_K -blk.10.ffn_gate_exps.weight iq3_xxs -blk.10.ffn_up_exps.weight iq3_xxs -blk.11.attn_k.weight iq3_xxs -blk.11.attn_q.weight iq3_xxs -blk.11.attn_v.weight q4_K -blk.11.ffn_gate_exps.weight iq3_xxs -blk.11.ffn_up_exps.weight iq3_xxs -blk.12.attn_k.weight iq3_xxs -blk.12.attn_q.weight iq3_xxs -blk.12.attn_v.weight q4_K -blk.12.ffn_gate_exps.weight iq3_xxs -blk.12.ffn_up_exps.weight iq3_xxs -blk.13.attn_k.weight iq3_xxs -blk.13.attn_q.weight iq3_xxs -blk.13.attn_v.weight q4_K -blk.13.ffn_gate_exps.weight iq3_xxs -blk.13.ffn_up_exps.weight iq3_xxs -blk.14.attn_k.weight iq3_xxs -blk.14.attn_q.weight iq3_xxs -blk.14.attn_v.weight q4_K -blk.14.ffn_gate_exps.weight iq3_xxs -blk.14.ffn_up_exps.weight iq3_xxs -blk.15.attn_k.weight iq3_xxs -blk.15.attn_q.weight iq3_xxs -blk.15.attn_v.weight q4_K -blk.15.ffn_gate_exps.weight iq3_xxs -blk.15.ffn_up_exps.weight iq3_xxs -blk.16.attn_k.weight iq3_xxs -blk.16.attn_q.weight iq3_xxs -blk.16.attn_v.weight q4_K -blk.16.ffn_gate_exps.weight iq3_xxs -blk.16.ffn_up_exps.weight iq3_xxs -blk.17.attn_k.weight iq3_xxs -blk.17.attn_q.weight iq3_xxs -blk.17.attn_v.weight q4_K -blk.17.ffn_gate_exps.weight iq3_xxs -blk.17.ffn_up_exps.weight iq3_xxs -blk.18.attn_k.weight iq3_xxs -blk.18.attn_q.weight iq3_xxs -blk.18.attn_v.weight q4_K -blk.18.ffn_gate_exps.weight iq3_xxs -blk.18.ffn_up_exps.weight iq3_xxs -blk.19.attn_k.weight iq3_xxs -blk.19.attn_q.weight iq3_xxs -blk.19.attn_v.weight q4_K -blk.19.ffn_gate_exps.weight iq3_xxs -blk.19.ffn_up_exps.weight iq3_xxs -blk.20.attn_k.weight iq3_xxs -blk.20.attn_q.weight iq3_xxs -blk.20.attn_v.weight q4_K -blk.20.ffn_gate_exps.weight iq3_xxs -blk.20.ffn_up_exps.weight iq3_xxs -blk.21.attn_k.weight iq3_xxs -blk.21.attn_q.weight iq3_xxs -blk.21.attn_v.weight q4_K -blk.21.ffn_gate_exps.weight iq3_xxs -blk.21.ffn_up_exps.weight iq3_xxs -blk.22.attn_k.weight iq3_xxs -blk.22.attn_q.weight iq3_xxs -blk.22.attn_v.weight q4_K -blk.22.ffn_gate_exps.weight iq3_xxs -blk.22.ffn_up_exps.weight iq3_xxs -blk.23.attn_k.weight iq3_xxs -blk.23.attn_q.weight iq3_xxs -blk.23.attn_v.weight q4_K -blk.23.ffn_gate_exps.weight iq3_xxs -blk.23.ffn_up_exps.weight iq3_xxs -blk.24.attn_k.weight iq3_xxs -blk.24.attn_q.weight iq3_xxs -blk.24.attn_v.weight q4_K -blk.24.ffn_gate_exps.weight iq3_xxs -blk.24.ffn_up_exps.weight iq3_xxs -blk.25.attn_k.weight iq3_xxs -blk.25.attn_q.weight iq3_xxs -blk.25.attn_v.weight q4_K -blk.25.ffn_gate_exps.weight iq3_xxs -blk.25.ffn_up_exps.weight iq3_xxs -blk.26.attn_k.weight iq3_xxs -blk.26.attn_q.weight iq3_xxs -blk.26.attn_v.weight q4_K -blk.26.ffn_gate_exps.weight iq3_xxs -blk.26.ffn_up_exps.weight iq3_xxs -blk.27.attn_k.weight iq3_xxs -blk.27.attn_q.weight iq3_xxs -blk.27.attn_v.weight q4_K -blk.27.ffn_gate_exps.weight iq3_xxs -blk.27.ffn_up_exps.weight iq3_xxs -blk.28.attn_k.weight iq3_xxs -blk.28.attn_q.weight iq3_xxs -blk.28.attn_v.weight q4_K -blk.28.ffn_gate_exps.weight iq3_xxs -blk.28.ffn_up_exps.weight iq3_xxs -blk.29.attn_k.weight iq3_xxs -blk.29.attn_q.weight iq3_xxs -blk.29.attn_v.weight q4_K -blk.29.ffn_gate_exps.weight iq3_xxs -blk.29.ffn_up_exps.weight iq3_xxs -blk.30.attn_k.weight iq3_xxs -blk.30.attn_q.weight iq3_xxs -blk.30.attn_v.weight q4_K -blk.30.ffn_gate_exps.weight iq3_xxs -blk.30.ffn_up_exps.weight iq3_xxs -blk.31.attn_k.weight iq3_xxs -blk.31.attn_q.weight iq3_xxs -blk.31.attn_v.weight q4_K -blk.32.attn_k.weight iq3_xxs -blk.32.attn_q.weight iq3_xxs -blk.32.attn_v.weight q4_K -blk.33.attn_k.weight iq3_xxs -blk.33.attn_q.weight iq3_xxs -blk.33.attn_v.weight q4_K -blk.34.attn_k.weight iq3_xxs -blk.34.attn_q.weight iq3_xxs -blk.34.attn_v.weight q4_K -blk.35.attn_k.weight iq3_xxs -blk.35.attn_q.weight iq3_xxs -blk.35.attn_v.weight q4_K +token_embd.weight iq4_nl +blk.0.attn_k.weight iq4_nl +blk.0.attn_q.weight iq4_nl +blk.0.attn_v.weight q5_0 +blk.0.ffn_down_exps.weight iq4_nl +blk.0.ffn_gate_exps.weight iq4_nl +blk.0.ffn_up_exps.weight iq4_nl +blk.1.attn_k.weight iq4_nl +blk.1.attn_q.weight iq4_nl +blk.1.attn_v.weight q5_0 +blk.1.ffn_down_exps.weight iq4_nl +blk.1.ffn_gate_exps.weight iq4_nl +blk.1.ffn_up_exps.weight iq4_nl +blk.2.attn_k.weight iq4_nl +blk.2.attn_q.weight iq4_nl +blk.2.attn_v.weight q5_0 +blk.2.ffn_down_exps.weight iq4_nl +blk.2.ffn_gate_exps.weight iq4_nl +blk.2.ffn_up_exps.weight iq4_nl +blk.3.attn_k.weight iq4_nl +blk.3.attn_q.weight iq4_nl +blk.3.attn_v.weight q5_0 +blk.3.ffn_down_exps.weight iq4_nl +blk.3.ffn_gate_exps.weight iq4_nl +blk.3.ffn_up_exps.weight iq4_nl +blk.4.attn_k.weight iq4_nl +blk.4.attn_q.weight iq4_nl +blk.4.attn_v.weight q5_0 +blk.4.ffn_down_exps.weight iq4_nl +blk.4.ffn_gate_exps.weight iq4_nl +blk.4.ffn_up_exps.weight iq4_nl +blk.5.attn_k.weight iq4_nl +blk.5.attn_q.weight iq4_nl +blk.5.attn_v.weight q5_0 +blk.5.ffn_down_exps.weight iq4_nl +blk.5.ffn_gate_exps.weight iq4_nl +blk.5.ffn_up_exps.weight iq4_nl +blk.6.attn_k.weight iq4_nl +blk.6.attn_q.weight iq4_nl +blk.6.attn_v.weight q5_0 +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_gate_exps.weight iq4_nl +blk.6.ffn_up_exps.weight iq4_nl +blk.7.attn_k.weight iq4_nl +blk.7.attn_q.weight iq4_nl +blk.7.attn_v.weight q5_0 +blk.7.ffn_down_exps.weight iq4_nl +blk.7.ffn_gate_exps.weight iq4_nl +blk.7.ffn_up_exps.weight iq4_nl +blk.8.attn_k.weight iq4_nl +blk.8.attn_q.weight iq4_nl +blk.8.attn_v.weight q5_0 +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_gate_exps.weight iq4_nl +blk.8.ffn_up_exps.weight iq4_nl +blk.9.attn_k.weight iq4_nl +blk.9.attn_q.weight iq4_nl +blk.9.attn_v.weight q5_0 +blk.9.ffn_down_exps.weight iq4_nl +blk.9.ffn_gate_exps.weight iq4_nl +blk.9.ffn_up_exps.weight iq4_nl +blk.10.attn_k.weight iq4_nl +blk.10.attn_q.weight iq4_nl +blk.10.attn_v.weight q5_0 +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_gate_exps.weight iq4_nl +blk.10.ffn_up_exps.weight iq4_nl +blk.11.attn_k.weight iq4_nl +blk.11.attn_q.weight iq4_nl +blk.11.attn_v.weight q5_0 +blk.11.ffn_down_exps.weight iq4_nl +blk.11.ffn_gate_exps.weight iq4_nl +blk.11.ffn_up_exps.weight iq4_nl +blk.12.attn_k.weight iq4_nl +blk.12.attn_q.weight iq4_nl +blk.12.attn_v.weight q5_0 +blk.12.ffn_down_exps.weight iq4_nl +blk.12.ffn_gate_exps.weight iq4_nl +blk.12.ffn_up_exps.weight iq4_nl +blk.13.attn_k.weight iq4_nl +blk.13.attn_q.weight iq4_nl +blk.13.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_gate_exps.weight iq4_nl +blk.13.ffn_up_exps.weight iq4_nl +blk.14.attn_k.weight iq4_nl +blk.14.attn_q.weight iq4_nl +blk.14.attn_v.weight q5_0 +blk.14.ffn_down_exps.weight iq4_nl +blk.14.ffn_gate_exps.weight iq4_nl +blk.14.ffn_up_exps.weight iq4_nl +blk.15.attn_k.weight iq4_nl +blk.15.attn_q.weight iq4_nl +blk.15.attn_v.weight q5_0 +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_gate_exps.weight iq4_nl +blk.15.ffn_up_exps.weight iq4_nl +blk.16.attn_k.weight iq4_nl +blk.16.attn_q.weight iq4_nl +blk.16.attn_v.weight q5_0 +blk.16.ffn_down_exps.weight iq4_nl +blk.16.ffn_gate_exps.weight iq4_nl +blk.16.ffn_up_exps.weight iq4_nl +blk.17.attn_k.weight iq4_nl +blk.17.attn_q.weight iq4_nl +blk.17.attn_v.weight q5_0 +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_gate_exps.weight iq4_nl +blk.17.ffn_up_exps.weight iq4_nl +blk.18.attn_k.weight iq4_nl +blk.18.attn_q.weight iq4_nl +blk.18.attn_v.weight q5_0 +blk.18.ffn_down_exps.weight iq4_nl +blk.18.ffn_gate_exps.weight iq4_nl +blk.18.ffn_up_exps.weight iq4_nl +blk.19.attn_k.weight iq4_nl +blk.19.attn_q.weight iq4_nl +blk.19.attn_v.weight q5_0 +blk.19.ffn_down_exps.weight iq4_nl +blk.19.ffn_gate_exps.weight iq4_nl +blk.19.ffn_up_exps.weight iq4_nl +blk.20.attn_k.weight iq4_nl +blk.20.attn_q.weight iq4_nl +blk.20.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_gate_exps.weight iq4_nl +blk.20.ffn_up_exps.weight iq4_nl +blk.21.attn_k.weight iq4_nl +blk.21.attn_q.weight iq4_nl +blk.21.attn_v.weight q5_0 +blk.21.ffn_down_exps.weight iq4_nl +blk.21.ffn_gate_exps.weight iq4_nl +blk.21.ffn_up_exps.weight iq4_nl +blk.22.attn_k.weight iq4_nl +blk.22.attn_q.weight iq4_nl +blk.22.attn_v.weight q5_0 +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_gate_exps.weight iq4_nl +blk.22.ffn_up_exps.weight iq4_nl +blk.23.attn_k.weight iq4_nl +blk.23.attn_q.weight iq4_nl +blk.23.attn_v.weight q5_0 +blk.23.ffn_down_exps.weight iq4_nl +blk.23.ffn_gate_exps.weight iq4_nl +blk.23.ffn_up_exps.weight iq4_nl +blk.24.attn_k.weight iq4_nl +blk.24.attn_q.weight iq4_nl +blk.24.attn_v.weight q5_0 +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_gate_exps.weight iq4_nl +blk.24.ffn_up_exps.weight iq4_nl +blk.25.attn_k.weight iq4_nl +blk.25.attn_q.weight iq4_nl +blk.25.attn_v.weight q5_0 +blk.25.ffn_down_exps.weight iq4_nl +blk.25.ffn_gate_exps.weight iq4_nl +blk.25.ffn_up_exps.weight iq4_nl +blk.26.attn_k.weight iq4_nl +blk.26.attn_q.weight iq4_nl +blk.26.attn_v.weight q5_0 +blk.26.ffn_down_exps.weight iq4_nl +blk.26.ffn_gate_exps.weight iq4_nl +blk.26.ffn_up_exps.weight iq4_nl +blk.27.attn_k.weight iq4_nl +blk.27.attn_q.weight iq4_nl +blk.27.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_gate_exps.weight iq4_nl +blk.27.ffn_up_exps.weight iq4_nl +blk.28.attn_k.weight iq4_nl +blk.28.attn_q.weight iq4_nl +blk.28.attn_v.weight q5_0 +blk.28.ffn_down_exps.weight iq4_nl +blk.28.ffn_gate_exps.weight iq4_nl +blk.28.ffn_up_exps.weight iq4_nl +blk.29.attn_k.weight iq4_nl +blk.29.attn_q.weight iq4_nl +blk.29.attn_v.weight q5_0 +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_gate_exps.weight iq4_nl +blk.29.ffn_up_exps.weight iq4_nl +blk.30.attn_k.weight iq4_nl +blk.30.attn_q.weight iq4_nl +blk.30.attn_v.weight q5_0 +blk.30.ffn_down_exps.weight iq4_nl +blk.30.ffn_gate_exps.weight iq4_nl +blk.30.ffn_up_exps.weight iq4_nl +blk.31.attn_k.weight iq4_nl +blk.31.attn_q.weight iq4_nl +blk.31.attn_v.weight q5_0 +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_gate_exps.weight iq4_nl +blk.31.ffn_up_exps.weight iq4_nl +blk.32.attn_k.weight iq4_nl +blk.32.attn_q.weight iq4_nl +blk.32.attn_v.weight q5_0 +blk.32.ffn_down_exps.weight iq4_nl +blk.32.ffn_gate_exps.weight iq4_nl +blk.32.ffn_up_exps.weight iq4_nl +blk.33.attn_k.weight iq4_nl +blk.33.attn_q.weight iq4_nl +blk.33.attn_v.weight q5_0 +blk.33.ffn_down_exps.weight iq4_nl +blk.33.ffn_gate_exps.weight iq4_nl +blk.33.ffn_up_exps.weight iq4_nl +blk.34.attn_k.weight iq4_nl +blk.34.attn_q.weight iq4_nl +blk.34.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_gate_exps.weight iq4_nl +blk.34.ffn_up_exps.weight iq4_nl +blk.35.attn_k.weight iq4_nl +blk.35.attn_q.weight iq4_nl +blk.35.attn_v.weight q5_0 +blk.35.ffn_down_exps.weight iq4_nl +blk.35.ffn_gate_exps.weight iq4_nl +blk.35.ffn_up_exps.weight iq4_nl [IQ3_XXS] iq3_xxs output.weight q8_0 -token_embd.weight iq3_s -blk.0.attn_k.weight iq2_s +token_embd.weight iq4_nl +blk.0.attn_k.weight iq4_nl blk.0.attn_output.weight iq3_s -blk.0.attn_q.weight iq2_s -blk.0.attn_v.weight q4_K -blk.0.ffn_down_exps.weight q4_K -blk.1.attn_k.weight iq2_s +blk.0.attn_q.weight iq4_nl +blk.0.attn_v.weight q5_0 +blk.0.ffn_down_exps.weight q5_0 +blk.0.ffn_gate_exps.weight iq4_nl +blk.0.ffn_up_exps.weight iq4_nl +blk.1.attn_k.weight iq4_nl blk.1.attn_output.weight iq3_s -blk.1.attn_q.weight iq2_s -blk.1.attn_v.weight q4_K -blk.1.ffn_down_exps.weight q4_K -blk.2.attn_k.weight iq2_s +blk.1.attn_q.weight iq4_nl +blk.1.attn_v.weight q5_0 +blk.1.ffn_down_exps.weight q5_0 +blk.1.ffn_gate_exps.weight iq4_nl +blk.1.ffn_up_exps.weight iq4_nl +blk.2.attn_k.weight iq4_nl blk.2.attn_output.weight iq3_s -blk.2.attn_q.weight iq2_s -blk.2.attn_v.weight q4_K -blk.2.ffn_down_exps.weight q4_K -blk.3.attn_k.weight iq2_s +blk.2.attn_q.weight iq4_nl +blk.2.attn_v.weight q5_0 +blk.2.ffn_down_exps.weight q5_0 +blk.2.ffn_gate_exps.weight iq4_nl +blk.2.ffn_up_exps.weight iq4_nl +blk.3.attn_k.weight iq4_nl blk.3.attn_output.weight iq3_s -blk.3.attn_q.weight iq2_s -blk.3.attn_v.weight q4_K -blk.3.ffn_down_exps.weight q4_K -blk.4.attn_k.weight iq2_s +blk.3.attn_q.weight iq4_nl +blk.3.attn_v.weight q5_0 +blk.3.ffn_down_exps.weight q5_0 +blk.3.ffn_gate_exps.weight iq4_nl +blk.3.ffn_up_exps.weight iq4_nl +blk.4.attn_k.weight iq4_nl blk.4.attn_output.weight iq3_s -blk.4.attn_q.weight iq2_s -blk.4.attn_v.weight q4_K -blk.4.ffn_down_exps.weight q3_K -blk.5.attn_k.weight iq2_s +blk.4.attn_q.weight iq4_nl +blk.4.attn_v.weight q5_0 +blk.4.ffn_down_exps.weight q4_0 +blk.4.ffn_gate_exps.weight iq4_nl +blk.4.ffn_up_exps.weight iq4_nl +blk.5.attn_k.weight iq4_nl blk.5.attn_output.weight iq3_s -blk.5.attn_q.weight iq2_s -blk.5.attn_v.weight q4_K -blk.5.ffn_down_exps.weight q3_K -blk.6.attn_k.weight iq2_s +blk.5.attn_q.weight iq4_nl +blk.5.attn_v.weight q5_0 +blk.5.ffn_down_exps.weight q4_0 +blk.5.ffn_gate_exps.weight iq4_nl +blk.5.ffn_up_exps.weight iq4_nl +blk.6.attn_k.weight iq4_nl blk.6.attn_output.weight iq3_s -blk.6.attn_q.weight iq2_s -blk.6.attn_v.weight q4_K -blk.6.ffn_down_exps.weight q3_K -blk.7.attn_k.weight iq2_s +blk.6.attn_q.weight iq4_nl +blk.6.attn_v.weight q5_0 +blk.6.ffn_down_exps.weight q4_0 +blk.6.ffn_gate_exps.weight iq4_nl +blk.6.ffn_up_exps.weight iq4_nl +blk.7.attn_k.weight iq4_nl blk.7.attn_output.weight iq3_s -blk.7.attn_q.weight iq2_s -blk.7.attn_v.weight q4_K -blk.7.ffn_down_exps.weight q3_K -blk.8.attn_k.weight iq2_s +blk.7.attn_q.weight iq4_nl +blk.7.attn_v.weight q5_0 +blk.7.ffn_down_exps.weight q4_0 +blk.7.ffn_gate_exps.weight iq4_nl +blk.7.ffn_up_exps.weight iq4_nl +blk.8.attn_k.weight iq4_nl blk.8.attn_output.weight iq3_s -blk.8.attn_q.weight iq2_s -blk.8.attn_v.weight q4_K -blk.8.ffn_down_exps.weight q3_K -blk.9.attn_k.weight iq2_s +blk.8.attn_q.weight iq4_nl +blk.8.attn_v.weight q5_0 +blk.8.ffn_down_exps.weight q4_0 +blk.8.ffn_gate_exps.weight iq4_nl +blk.8.ffn_up_exps.weight iq4_nl +blk.9.attn_k.weight iq4_nl blk.9.attn_output.weight iq3_s -blk.9.attn_q.weight iq2_s -blk.9.attn_v.weight q4_K -blk.9.ffn_down_exps.weight q3_K -blk.10.attn_k.weight iq2_s +blk.9.attn_q.weight iq4_nl +blk.9.attn_v.weight q5_0 +blk.9.ffn_down_exps.weight q4_0 +blk.9.ffn_gate_exps.weight iq4_nl +blk.9.ffn_up_exps.weight iq4_nl +blk.10.attn_k.weight iq4_nl blk.10.attn_output.weight iq3_s -blk.10.attn_q.weight iq2_s -blk.10.attn_v.weight q4_K -blk.10.ffn_down_exps.weight q3_K -blk.11.attn_k.weight iq2_s +blk.10.attn_q.weight iq4_nl +blk.10.attn_v.weight q5_0 +blk.10.ffn_down_exps.weight q4_0 +blk.10.ffn_gate_exps.weight iq4_nl +blk.10.ffn_up_exps.weight iq4_nl +blk.11.attn_k.weight iq4_nl blk.11.attn_output.weight iq3_s -blk.11.attn_q.weight iq2_s -blk.11.attn_v.weight q4_K -blk.11.ffn_down_exps.weight q3_K -blk.12.attn_k.weight iq2_s +blk.11.attn_q.weight iq4_nl +blk.11.attn_v.weight q5_0 +blk.11.ffn_down_exps.weight q4_0 +blk.11.ffn_gate_exps.weight iq4_nl +blk.11.ffn_up_exps.weight iq4_nl +blk.12.attn_k.weight iq4_nl blk.12.attn_output.weight iq3_s -blk.12.attn_q.weight iq2_s -blk.12.attn_v.weight q4_K -blk.12.ffn_down_exps.weight q3_K -blk.13.attn_k.weight iq2_s +blk.12.attn_q.weight iq4_nl +blk.12.attn_v.weight q5_0 +blk.12.ffn_down_exps.weight q4_0 +blk.12.ffn_gate_exps.weight iq4_nl +blk.12.ffn_up_exps.weight iq4_nl +blk.13.attn_k.weight iq4_nl blk.13.attn_output.weight iq3_s -blk.13.attn_q.weight iq2_s -blk.13.attn_v.weight q4_K -blk.13.ffn_down_exps.weight q3_K -blk.14.attn_k.weight iq2_s +blk.13.attn_q.weight iq4_nl +blk.13.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight q4_0 +blk.13.ffn_gate_exps.weight iq4_nl +blk.13.ffn_up_exps.weight iq4_nl +blk.14.attn_k.weight iq4_nl blk.14.attn_output.weight iq3_s -blk.14.attn_q.weight iq2_s -blk.14.attn_v.weight q4_K -blk.14.ffn_down_exps.weight q3_K -blk.15.attn_k.weight iq2_s +blk.14.attn_q.weight iq4_nl +blk.14.attn_v.weight q5_0 +blk.14.ffn_down_exps.weight q4_0 +blk.14.ffn_gate_exps.weight iq4_nl +blk.14.ffn_up_exps.weight iq4_nl +blk.15.attn_k.weight iq4_nl blk.15.attn_output.weight iq3_s -blk.15.attn_q.weight iq2_s -blk.15.attn_v.weight q4_K -blk.15.ffn_down_exps.weight q3_K -blk.16.attn_k.weight iq2_s +blk.15.attn_q.weight iq4_nl +blk.15.attn_v.weight q5_0 +blk.15.ffn_down_exps.weight q4_0 +blk.15.ffn_gate_exps.weight iq4_nl +blk.15.ffn_up_exps.weight iq4_nl +blk.16.attn_k.weight iq4_nl blk.16.attn_output.weight iq3_s -blk.16.attn_q.weight iq2_s -blk.16.attn_v.weight q4_K -blk.16.ffn_down_exps.weight q3_K -blk.17.attn_k.weight iq2_s +blk.16.attn_q.weight iq4_nl +blk.16.attn_v.weight q5_0 +blk.16.ffn_down_exps.weight q4_0 +blk.16.ffn_gate_exps.weight iq4_nl +blk.16.ffn_up_exps.weight iq4_nl +blk.17.attn_k.weight iq4_nl blk.17.attn_output.weight iq3_s -blk.17.attn_q.weight iq2_s -blk.17.attn_v.weight q4_K -blk.17.ffn_down_exps.weight q3_K -blk.18.attn_k.weight iq2_s +blk.17.attn_q.weight iq4_nl +blk.17.attn_v.weight q5_0 +blk.17.ffn_down_exps.weight q4_0 +blk.17.ffn_gate_exps.weight iq4_nl +blk.17.ffn_up_exps.weight iq4_nl +blk.18.attn_k.weight iq4_nl blk.18.attn_output.weight iq3_s -blk.18.attn_q.weight iq2_s -blk.18.attn_v.weight q4_K -blk.18.ffn_down_exps.weight q3_K -blk.19.attn_k.weight iq2_s +blk.18.attn_q.weight iq4_nl +blk.18.attn_v.weight q5_0 +blk.18.ffn_down_exps.weight q4_0 +blk.18.ffn_gate_exps.weight iq4_nl +blk.18.ffn_up_exps.weight iq4_nl +blk.19.attn_k.weight iq4_nl blk.19.attn_output.weight iq3_s -blk.19.attn_q.weight iq2_s -blk.19.attn_v.weight q4_K -blk.19.ffn_down_exps.weight q3_K -blk.20.attn_k.weight iq2_s +blk.19.attn_q.weight iq4_nl +blk.19.attn_v.weight q5_0 +blk.19.ffn_down_exps.weight q4_0 +blk.19.ffn_gate_exps.weight iq4_nl +blk.19.ffn_up_exps.weight iq4_nl +blk.20.attn_k.weight iq4_nl blk.20.attn_output.weight iq3_s -blk.20.attn_q.weight iq2_s -blk.20.attn_v.weight q4_K -blk.20.ffn_down_exps.weight q3_K -blk.21.attn_k.weight iq2_s +blk.20.attn_q.weight iq4_nl +blk.20.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight q4_0 +blk.20.ffn_gate_exps.weight iq4_nl +blk.20.ffn_up_exps.weight iq4_nl +blk.21.attn_k.weight iq4_nl blk.21.attn_output.weight iq3_s -blk.21.attn_q.weight iq2_s -blk.21.attn_v.weight q4_K -blk.21.ffn_down_exps.weight q3_K -blk.22.attn_k.weight iq2_s +blk.21.attn_q.weight iq4_nl +blk.21.attn_v.weight q5_0 +blk.21.ffn_down_exps.weight q4_0 +blk.21.ffn_gate_exps.weight iq4_nl +blk.21.ffn_up_exps.weight iq4_nl +blk.22.attn_k.weight iq4_nl blk.22.attn_output.weight iq3_s -blk.22.attn_q.weight iq2_s -blk.22.attn_v.weight q4_K -blk.22.ffn_down_exps.weight q3_K -blk.23.attn_k.weight iq2_s +blk.22.attn_q.weight iq4_nl +blk.22.attn_v.weight q5_0 +blk.22.ffn_down_exps.weight q4_0 +blk.22.ffn_gate_exps.weight iq4_nl +blk.22.ffn_up_exps.weight iq4_nl +blk.23.attn_k.weight iq4_nl blk.23.attn_output.weight iq3_s -blk.23.attn_q.weight iq2_s -blk.23.attn_v.weight q4_K -blk.23.ffn_down_exps.weight q3_K -blk.24.attn_k.weight iq2_s +blk.23.attn_q.weight iq4_nl +blk.23.attn_v.weight q5_0 +blk.23.ffn_down_exps.weight q4_0 +blk.23.ffn_gate_exps.weight iq4_nl +blk.23.ffn_up_exps.weight iq4_nl +blk.24.attn_k.weight iq4_nl blk.24.attn_output.weight iq3_s -blk.24.attn_q.weight iq2_s -blk.24.attn_v.weight q4_K -blk.24.ffn_down_exps.weight q3_K -blk.25.attn_k.weight iq2_s +blk.24.attn_q.weight iq4_nl +blk.24.attn_v.weight q5_0 +blk.24.ffn_down_exps.weight q4_0 +blk.24.ffn_gate_exps.weight iq4_nl +blk.24.ffn_up_exps.weight iq4_nl +blk.25.attn_k.weight iq4_nl blk.25.attn_output.weight iq3_s -blk.25.attn_q.weight iq2_s -blk.25.attn_v.weight q4_K -blk.25.ffn_down_exps.weight q3_K -blk.26.attn_k.weight iq2_s +blk.25.attn_q.weight iq4_nl +blk.25.attn_v.weight q5_0 +blk.25.ffn_down_exps.weight q4_0 +blk.25.ffn_gate_exps.weight iq4_nl +blk.25.ffn_up_exps.weight iq4_nl +blk.26.attn_k.weight iq4_nl blk.26.attn_output.weight iq3_s -blk.26.attn_q.weight iq2_s -blk.26.attn_v.weight q4_K -blk.26.ffn_down_exps.weight q3_K -blk.27.attn_k.weight iq2_s +blk.26.attn_q.weight iq4_nl +blk.26.attn_v.weight q5_0 +blk.26.ffn_down_exps.weight q4_0 +blk.26.ffn_gate_exps.weight iq4_nl +blk.26.ffn_up_exps.weight iq4_nl +blk.27.attn_k.weight iq4_nl blk.27.attn_output.weight iq3_s -blk.27.attn_q.weight iq2_s -blk.27.attn_v.weight q4_K -blk.27.ffn_down_exps.weight q3_K -blk.28.attn_k.weight iq2_s +blk.27.attn_q.weight iq4_nl +blk.27.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight q4_0 +blk.27.ffn_gate_exps.weight iq4_nl +blk.27.ffn_up_exps.weight iq4_nl +blk.28.attn_k.weight iq4_nl blk.28.attn_output.weight iq3_s -blk.28.attn_q.weight iq2_s -blk.28.attn_v.weight q4_K -blk.28.ffn_down_exps.weight q3_K -blk.29.attn_k.weight iq2_s +blk.28.attn_q.weight iq4_nl +blk.28.attn_v.weight q5_0 +blk.28.ffn_down_exps.weight q4_0 +blk.28.ffn_gate_exps.weight iq4_nl +blk.28.ffn_up_exps.weight iq4_nl +blk.29.attn_k.weight iq4_nl blk.29.attn_output.weight iq3_s -blk.29.attn_q.weight iq2_s -blk.29.attn_v.weight q4_K -blk.29.ffn_down_exps.weight q3_K -blk.30.attn_k.weight iq2_s +blk.29.attn_q.weight iq4_nl +blk.29.attn_v.weight q5_0 +blk.29.ffn_down_exps.weight q4_0 +blk.29.ffn_gate_exps.weight iq4_nl +blk.29.ffn_up_exps.weight iq4_nl +blk.30.attn_k.weight iq4_nl blk.30.attn_output.weight iq3_s -blk.30.attn_q.weight iq2_s -blk.30.attn_v.weight q4_K -blk.30.ffn_down_exps.weight q3_K -blk.31.attn_k.weight iq2_s +blk.30.attn_q.weight iq4_nl +blk.30.attn_v.weight q5_0 +blk.30.ffn_down_exps.weight q4_0 +blk.30.ffn_gate_exps.weight iq4_nl +blk.30.ffn_up_exps.weight iq4_nl +blk.31.attn_k.weight iq4_nl blk.31.attn_output.weight iq3_s -blk.31.attn_q.weight iq2_s -blk.31.attn_v.weight q4_K -blk.31.ffn_down_exps.weight q3_K -blk.32.attn_k.weight iq2_s +blk.31.attn_q.weight iq4_nl +blk.31.attn_v.weight q5_0 +blk.31.ffn_down_exps.weight q4_0 +blk.31.ffn_gate_exps.weight iq4_nl +blk.31.ffn_up_exps.weight iq4_nl +blk.32.attn_k.weight iq4_nl blk.32.attn_output.weight iq3_s -blk.32.attn_q.weight iq2_s -blk.32.attn_v.weight q4_K -blk.32.ffn_down_exps.weight q3_K -blk.33.attn_k.weight iq2_s +blk.32.attn_q.weight iq4_nl +blk.32.attn_v.weight q5_0 +blk.32.ffn_down_exps.weight q4_0 +blk.32.ffn_gate_exps.weight iq4_nl +blk.32.ffn_up_exps.weight iq4_nl +blk.33.attn_k.weight iq4_nl blk.33.attn_output.weight iq3_s -blk.33.attn_q.weight iq2_s -blk.33.attn_v.weight q4_K -blk.33.ffn_down_exps.weight q3_K -blk.34.attn_k.weight iq2_s +blk.33.attn_q.weight iq4_nl +blk.33.attn_v.weight q5_0 +blk.33.ffn_down_exps.weight q4_0 +blk.33.ffn_gate_exps.weight iq4_nl +blk.33.ffn_up_exps.weight iq4_nl +blk.34.attn_k.weight iq4_nl blk.34.attn_output.weight iq3_s -blk.34.attn_q.weight iq2_s -blk.34.attn_v.weight q4_K -blk.34.ffn_down_exps.weight q3_K -blk.35.attn_k.weight iq2_s +blk.34.attn_q.weight iq4_nl +blk.34.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight q4_0 +blk.34.ffn_gate_exps.weight iq4_nl +blk.34.ffn_up_exps.weight iq4_nl +blk.35.attn_k.weight iq4_nl blk.35.attn_output.weight iq3_s -blk.35.attn_q.weight iq2_s -blk.35.attn_v.weight q4_K -blk.35.ffn_down_exps.weight q3_K +blk.35.attn_q.weight iq4_nl +blk.35.attn_v.weight q5_0 +blk.35.ffn_down_exps.weight q4_0 +blk.35.ffn_gate_exps.weight iq4_nl +blk.35.ffn_up_exps.weight iq4_nl [IQ1_S] iq1_s output.weight q8_0 -token_embd.weight q2_K +token_embd.weight q4_0 +blk.0.attn_k.weight iq4_nl blk.0.attn_output.weight iq2_xxs -blk.0.attn_v.weight q4_K -blk.0.ffn_down_exps.weight q2_K +blk.0.attn_q.weight iq4_nl +blk.0.attn_v.weight q5_0 +blk.0.ffn_down_exps.weight q4_0 +blk.0.ffn_gate_exps.weight iq4_nl +blk.0.ffn_up_exps.weight iq4_nl +blk.1.attn_k.weight iq4_nl blk.1.attn_output.weight iq2_xxs -blk.1.attn_v.weight q4_K -blk.1.ffn_down_exps.weight q2_K +blk.1.attn_q.weight iq4_nl +blk.1.attn_v.weight q5_0 +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_gate_exps.weight iq4_nl +blk.1.ffn_up_exps.weight iq4_nl +blk.2.attn_k.weight iq4_nl blk.2.attn_output.weight iq2_xxs -blk.2.attn_v.weight q4_K -blk.2.ffn_down_exps.weight q2_K +blk.2.attn_q.weight iq4_nl +blk.2.attn_v.weight q5_0 +blk.2.ffn_down_exps.weight q4_0 +blk.2.ffn_gate_exps.weight iq4_nl +blk.2.ffn_up_exps.weight iq4_nl +blk.3.attn_k.weight iq4_nl blk.3.attn_output.weight iq2_xxs -blk.3.attn_v.weight q4_K -blk.3.ffn_down_exps.weight q2_K +blk.3.attn_q.weight iq4_nl +blk.3.attn_v.weight q5_0 +blk.3.ffn_down_exps.weight q4_0 +blk.3.ffn_gate_exps.weight iq4_nl +blk.3.ffn_up_exps.weight iq4_nl +blk.4.attn_k.weight iq4_nl blk.4.attn_output.weight iq2_xxs -blk.4.attn_v.weight q4_K +blk.4.attn_q.weight iq4_nl +blk.4.attn_v.weight q5_0 +blk.4.ffn_down_exps.weight iq4_nl +blk.4.ffn_gate_exps.weight iq4_nl +blk.4.ffn_up_exps.weight iq4_nl +blk.5.attn_k.weight iq4_nl blk.5.attn_output.weight iq2_xxs -blk.5.attn_v.weight q4_K +blk.5.attn_q.weight iq4_nl +blk.5.attn_v.weight q5_0 +blk.5.ffn_down_exps.weight iq4_nl +blk.5.ffn_gate_exps.weight iq4_nl +blk.5.ffn_up_exps.weight iq4_nl +blk.6.attn_k.weight iq4_nl blk.6.attn_output.weight iq2_xxs -blk.6.attn_v.weight q4_K +blk.6.attn_q.weight iq4_nl +blk.6.attn_v.weight q5_0 +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_gate_exps.weight iq4_nl +blk.6.ffn_up_exps.weight iq4_nl +blk.7.attn_k.weight iq4_nl blk.7.attn_output.weight iq2_xxs -blk.7.attn_v.weight q4_K +blk.7.attn_q.weight iq4_nl +blk.7.attn_v.weight q5_0 +blk.7.ffn_down_exps.weight iq4_nl +blk.7.ffn_gate_exps.weight iq4_nl +blk.7.ffn_up_exps.weight iq4_nl +blk.8.attn_k.weight iq4_nl blk.8.attn_output.weight iq2_xxs -blk.8.attn_v.weight q4_K +blk.8.attn_q.weight iq4_nl +blk.8.attn_v.weight q5_0 +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_gate_exps.weight iq4_nl +blk.8.ffn_up_exps.weight iq4_nl +blk.9.attn_k.weight iq4_nl blk.9.attn_output.weight iq2_xxs -blk.9.attn_v.weight q4_K +blk.9.attn_q.weight iq4_nl +blk.9.attn_v.weight q5_0 +blk.9.ffn_down_exps.weight iq4_nl +blk.9.ffn_gate_exps.weight iq4_nl +blk.9.ffn_up_exps.weight iq4_nl +blk.10.attn_k.weight iq4_nl blk.10.attn_output.weight iq2_xxs -blk.10.attn_v.weight q4_K +blk.10.attn_q.weight iq4_nl +blk.10.attn_v.weight q5_0 +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_gate_exps.weight iq4_nl +blk.10.ffn_up_exps.weight iq4_nl +blk.11.attn_k.weight iq4_nl blk.11.attn_output.weight iq2_xxs -blk.11.attn_v.weight q4_K +blk.11.attn_q.weight iq4_nl +blk.11.attn_v.weight q5_0 +blk.11.ffn_down_exps.weight iq4_nl +blk.11.ffn_gate_exps.weight iq4_nl +blk.11.ffn_up_exps.weight iq4_nl +blk.12.attn_k.weight iq4_nl blk.12.attn_output.weight iq2_xxs -blk.12.attn_v.weight q4_K +blk.12.attn_q.weight iq4_nl +blk.12.attn_v.weight q5_0 +blk.12.ffn_down_exps.weight iq4_nl +blk.12.ffn_gate_exps.weight iq4_nl +blk.12.ffn_up_exps.weight iq4_nl +blk.13.attn_k.weight iq4_nl blk.13.attn_output.weight iq2_xxs -blk.13.attn_v.weight q4_K +blk.13.attn_q.weight iq4_nl +blk.13.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_gate_exps.weight iq4_nl +blk.13.ffn_up_exps.weight iq4_nl +blk.14.attn_k.weight iq4_nl blk.14.attn_output.weight iq2_xxs -blk.14.attn_v.weight q4_K +blk.14.attn_q.weight iq4_nl +blk.14.attn_v.weight q5_0 +blk.14.ffn_down_exps.weight iq4_nl +blk.14.ffn_gate_exps.weight iq4_nl +blk.14.ffn_up_exps.weight iq4_nl +blk.15.attn_k.weight iq4_nl blk.15.attn_output.weight iq2_xxs -blk.15.attn_v.weight q4_K +blk.15.attn_q.weight iq4_nl +blk.15.attn_v.weight q5_0 +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_gate_exps.weight iq4_nl +blk.15.ffn_up_exps.weight iq4_nl +blk.16.attn_k.weight iq4_nl blk.16.attn_output.weight iq2_xxs -blk.16.attn_v.weight q4_K +blk.16.attn_q.weight iq4_nl +blk.16.attn_v.weight q5_0 +blk.16.ffn_down_exps.weight iq4_nl +blk.16.ffn_gate_exps.weight iq4_nl +blk.16.ffn_up_exps.weight iq4_nl +blk.17.attn_k.weight iq4_nl blk.17.attn_output.weight iq2_xxs -blk.17.attn_v.weight q4_K +blk.17.attn_q.weight iq4_nl +blk.17.attn_v.weight q5_0 +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_gate_exps.weight iq4_nl +blk.17.ffn_up_exps.weight iq4_nl +blk.18.attn_k.weight iq4_nl blk.18.attn_output.weight iq2_xxs -blk.18.attn_v.weight q4_K +blk.18.attn_q.weight iq4_nl +blk.18.attn_v.weight q5_0 +blk.18.ffn_down_exps.weight iq4_nl +blk.18.ffn_gate_exps.weight iq4_nl +blk.18.ffn_up_exps.weight iq4_nl +blk.19.attn_k.weight iq4_nl blk.19.attn_output.weight iq2_xxs -blk.19.attn_v.weight q4_K +blk.19.attn_q.weight iq4_nl +blk.19.attn_v.weight q5_0 +blk.19.ffn_down_exps.weight iq4_nl +blk.19.ffn_gate_exps.weight iq4_nl +blk.19.ffn_up_exps.weight iq4_nl +blk.20.attn_k.weight iq4_nl blk.20.attn_output.weight iq2_xxs -blk.20.attn_v.weight q4_K +blk.20.attn_q.weight iq4_nl +blk.20.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_gate_exps.weight iq4_nl +blk.20.ffn_up_exps.weight iq4_nl +blk.21.attn_k.weight iq4_nl blk.21.attn_output.weight iq2_xxs -blk.21.attn_v.weight q4_K +blk.21.attn_q.weight iq4_nl +blk.21.attn_v.weight q5_0 +blk.21.ffn_down_exps.weight iq4_nl +blk.21.ffn_gate_exps.weight iq4_nl +blk.21.ffn_up_exps.weight iq4_nl +blk.22.attn_k.weight iq4_nl blk.22.attn_output.weight iq2_xxs -blk.22.attn_v.weight q4_K +blk.22.attn_q.weight iq4_nl +blk.22.attn_v.weight q5_0 +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_gate_exps.weight iq4_nl +blk.22.ffn_up_exps.weight iq4_nl +blk.23.attn_k.weight iq4_nl blk.23.attn_output.weight iq2_xxs -blk.23.attn_v.weight q4_K +blk.23.attn_q.weight iq4_nl +blk.23.attn_v.weight q5_0 +blk.23.ffn_down_exps.weight iq4_nl +blk.23.ffn_gate_exps.weight iq4_nl +blk.23.ffn_up_exps.weight iq4_nl +blk.24.attn_k.weight iq4_nl blk.24.attn_output.weight iq2_xxs -blk.24.attn_v.weight q4_K +blk.24.attn_q.weight iq4_nl +blk.24.attn_v.weight q5_0 +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_gate_exps.weight iq4_nl +blk.24.ffn_up_exps.weight iq4_nl +blk.25.attn_k.weight iq4_nl blk.25.attn_output.weight iq2_xxs -blk.25.attn_v.weight q4_K +blk.25.attn_q.weight iq4_nl +blk.25.attn_v.weight q5_0 +blk.25.ffn_down_exps.weight iq4_nl +blk.25.ffn_gate_exps.weight iq4_nl +blk.25.ffn_up_exps.weight iq4_nl +blk.26.attn_k.weight iq4_nl blk.26.attn_output.weight iq2_xxs -blk.26.attn_v.weight q4_K +blk.26.attn_q.weight iq4_nl +blk.26.attn_v.weight q5_0 +blk.26.ffn_down_exps.weight iq4_nl +blk.26.ffn_gate_exps.weight iq4_nl +blk.26.ffn_up_exps.weight iq4_nl +blk.27.attn_k.weight iq4_nl blk.27.attn_output.weight iq2_xxs -blk.27.attn_v.weight q4_K +blk.27.attn_q.weight iq4_nl +blk.27.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_gate_exps.weight iq4_nl +blk.27.ffn_up_exps.weight iq4_nl +blk.28.attn_k.weight iq4_nl blk.28.attn_output.weight iq2_xxs -blk.28.attn_v.weight q4_K +blk.28.attn_q.weight iq4_nl +blk.28.attn_v.weight q5_0 +blk.28.ffn_down_exps.weight iq4_nl +blk.28.ffn_gate_exps.weight iq4_nl +blk.28.ffn_up_exps.weight iq4_nl +blk.29.attn_k.weight iq4_nl blk.29.attn_output.weight iq2_xxs -blk.29.attn_v.weight q4_K +blk.29.attn_q.weight iq4_nl +blk.29.attn_v.weight q5_0 +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_gate_exps.weight iq4_nl +blk.29.ffn_up_exps.weight iq4_nl +blk.30.attn_k.weight iq4_nl blk.30.attn_output.weight iq2_xxs -blk.30.attn_v.weight q4_K +blk.30.attn_q.weight iq4_nl +blk.30.attn_v.weight q5_0 +blk.30.ffn_down_exps.weight iq4_nl +blk.30.ffn_gate_exps.weight iq4_nl +blk.30.ffn_up_exps.weight iq4_nl +blk.31.attn_k.weight iq4_nl blk.31.attn_output.weight iq2_xxs -blk.31.attn_v.weight q4_K +blk.31.attn_q.weight iq4_nl +blk.31.attn_v.weight q5_0 +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_gate_exps.weight iq4_nl +blk.31.ffn_up_exps.weight iq4_nl +blk.32.attn_k.weight iq4_nl blk.32.attn_output.weight iq2_xxs -blk.32.attn_v.weight q4_K +blk.32.attn_q.weight iq4_nl +blk.32.attn_v.weight q5_0 +blk.32.ffn_down_exps.weight iq4_nl +blk.32.ffn_gate_exps.weight iq4_nl +blk.32.ffn_up_exps.weight iq4_nl +blk.33.attn_k.weight iq4_nl blk.33.attn_output.weight iq2_xxs -blk.33.attn_v.weight q4_K +blk.33.attn_q.weight iq4_nl +blk.33.attn_v.weight q5_0 +blk.33.ffn_down_exps.weight iq4_nl +blk.33.ffn_gate_exps.weight iq4_nl +blk.33.ffn_up_exps.weight iq4_nl +blk.34.attn_k.weight iq4_nl blk.34.attn_output.weight iq2_xxs -blk.34.attn_v.weight q4_K +blk.34.attn_q.weight iq4_nl +blk.34.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_gate_exps.weight iq4_nl +blk.34.ffn_up_exps.weight iq4_nl +blk.35.attn_k.weight iq4_nl blk.35.attn_output.weight iq2_xxs -blk.35.attn_v.weight q4_K +blk.35.attn_q.weight iq4_nl +blk.35.attn_v.weight q5_0 +blk.35.ffn_down_exps.weight iq4_nl +blk.35.ffn_gate_exps.weight iq4_nl +blk.35.ffn_up_exps.weight iq4_nl [IQ4_NL] iq4_nl -output.weight q6_K -blk.0.attn_v.weight q5_K -blk.0.ffn_down_exps.weight q5_K -blk.1.attn_v.weight q5_K -blk.1.ffn_down_exps.weight q5_K -blk.2.attn_v.weight q5_K -blk.2.ffn_down_exps.weight q5_K -blk.3.attn_v.weight q5_K -blk.3.ffn_down_exps.weight q5_K -blk.4.attn_v.weight q5_K -blk.5.attn_v.weight q5_K -blk.6.attn_v.weight q5_K -blk.7.attn_v.weight q5_K -blk.8.attn_v.weight q5_K -blk.9.attn_v.weight q5_K -blk.10.attn_v.weight q5_K -blk.11.attn_v.weight q5_K -blk.12.attn_v.weight q5_K -blk.13.attn_v.weight q5_K -blk.14.attn_v.weight q5_K -blk.15.attn_v.weight q5_K -blk.16.attn_v.weight q5_K -blk.17.attn_v.weight q5_K -blk.18.attn_v.weight q5_K -blk.19.attn_v.weight q5_K -blk.20.attn_v.weight q5_K -blk.21.attn_v.weight q5_K -blk.22.attn_v.weight q5_K -blk.23.attn_v.weight q5_K -blk.24.attn_v.weight q5_K -blk.25.attn_v.weight q5_K -blk.26.attn_v.weight q5_K -blk.27.attn_v.weight q5_K -blk.28.attn_v.weight q5_K -blk.29.attn_v.weight q5_K -blk.30.attn_v.weight q5_K -blk.31.attn_v.weight q5_K -blk.32.attn_v.weight q5_K -blk.33.attn_v.weight q5_K -blk.34.attn_v.weight q5_K -blk.35.attn_v.weight q5_K +output.weight q8_0 +blk.0.attn_v.weight q5_1 +blk.0.ffn_down_exps.weight q5_1 +blk.1.attn_v.weight q5_1 +blk.1.ffn_down_exps.weight q5_1 +blk.2.attn_v.weight q5_1 +blk.2.ffn_down_exps.weight q5_1 +blk.3.attn_v.weight q5_1 +blk.3.ffn_down_exps.weight q5_1 +blk.4.attn_v.weight q5_1 +blk.5.attn_v.weight q5_1 +blk.6.attn_v.weight q5_1 +blk.7.attn_v.weight q5_1 +blk.8.attn_v.weight q5_1 +blk.9.attn_v.weight q5_1 +blk.10.attn_v.weight q5_1 +blk.11.attn_v.weight q5_1 +blk.12.attn_v.weight q5_1 +blk.13.attn_v.weight q5_1 +blk.14.attn_v.weight q5_1 +blk.15.attn_v.weight q5_1 +blk.16.attn_v.weight q5_1 +blk.17.attn_v.weight q5_1 +blk.18.attn_v.weight q5_1 +blk.19.attn_v.weight q5_1 +blk.20.attn_v.weight q5_1 +blk.21.attn_v.weight q5_1 +blk.22.attn_v.weight q5_1 +blk.23.attn_v.weight q5_1 +blk.24.attn_v.weight q5_1 +blk.25.attn_v.weight q5_1 +blk.26.attn_v.weight q5_1 +blk.27.attn_v.weight q5_1 +blk.28.attn_v.weight q5_1 +blk.29.attn_v.weight q5_1 +blk.30.attn_v.weight q5_1 +blk.31.attn_v.weight q5_1 +blk.32.attn_v.weight q5_1 +blk.33.attn_v.weight q5_1 +blk.34.attn_v.weight q5_1 +blk.35.attn_v.weight q5_1 [IQ3_S] iq3_s output.weight q8_0 -blk.0.attn_v.weight q4_K -blk.1.attn_v.weight q4_K -blk.2.attn_v.weight q4_K -blk.3.attn_v.weight q4_K -blk.4.attn_v.weight q4_K -blk.5.attn_v.weight q4_K -blk.6.attn_v.weight q4_K -blk.7.attn_v.weight q4_K -blk.8.attn_v.weight q4_K -blk.9.attn_v.weight q4_K -blk.10.attn_v.weight q4_K -blk.11.attn_v.weight q4_K -blk.12.attn_v.weight q4_K -blk.13.attn_v.weight q4_K -blk.14.attn_v.weight q4_K -blk.15.attn_v.weight q4_K -blk.16.attn_v.weight q4_K -blk.17.attn_v.weight q4_K -blk.18.attn_v.weight q4_K -blk.19.attn_v.weight q4_K -blk.20.attn_v.weight q4_K -blk.21.attn_v.weight q4_K -blk.22.attn_v.weight q4_K -blk.23.attn_v.weight q4_K -blk.24.attn_v.weight q4_K -blk.25.attn_v.weight q4_K -blk.26.attn_v.weight q4_K -blk.27.attn_v.weight q4_K -blk.28.attn_v.weight q4_K -blk.29.attn_v.weight q4_K -blk.30.attn_v.weight q4_K -blk.31.attn_v.weight q4_K -blk.32.attn_v.weight q4_K -blk.33.attn_v.weight q4_K -blk.34.attn_v.weight q4_K -blk.35.attn_v.weight q4_K +token_embd.weight iq4_nl +blk.0.attn_k.weight iq4_nl +blk.0.attn_q.weight iq4_nl +blk.0.attn_v.weight q5_0 +blk.0.ffn_down_exps.weight iq4_nl +blk.0.ffn_gate_exps.weight iq4_nl +blk.0.ffn_up_exps.weight iq4_nl +blk.1.attn_k.weight iq4_nl +blk.1.attn_q.weight iq4_nl +blk.1.attn_v.weight q5_0 +blk.1.ffn_down_exps.weight iq4_nl +blk.1.ffn_gate_exps.weight iq4_nl +blk.1.ffn_up_exps.weight iq4_nl +blk.2.attn_k.weight iq4_nl +blk.2.attn_q.weight iq4_nl +blk.2.attn_v.weight q5_0 +blk.2.ffn_down_exps.weight iq4_nl +blk.2.ffn_gate_exps.weight iq4_nl +blk.2.ffn_up_exps.weight iq4_nl +blk.3.attn_k.weight iq4_nl +blk.3.attn_q.weight iq4_nl +blk.3.attn_v.weight q5_0 +blk.3.ffn_down_exps.weight iq4_nl +blk.3.ffn_gate_exps.weight iq4_nl +blk.3.ffn_up_exps.weight iq4_nl +blk.4.attn_k.weight iq4_nl +blk.4.attn_q.weight iq4_nl +blk.4.attn_v.weight q5_0 +blk.4.ffn_down_exps.weight iq4_nl +blk.4.ffn_gate_exps.weight iq4_nl +blk.4.ffn_up_exps.weight iq4_nl +blk.5.attn_k.weight iq4_nl +blk.5.attn_q.weight iq4_nl +blk.5.attn_v.weight q5_0 +blk.5.ffn_down_exps.weight iq4_nl +blk.5.ffn_gate_exps.weight iq4_nl +blk.5.ffn_up_exps.weight iq4_nl +blk.6.attn_k.weight iq4_nl +blk.6.attn_q.weight iq4_nl +blk.6.attn_v.weight q5_0 +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_gate_exps.weight iq4_nl +blk.6.ffn_up_exps.weight iq4_nl +blk.7.attn_k.weight iq4_nl +blk.7.attn_q.weight iq4_nl +blk.7.attn_v.weight q5_0 +blk.7.ffn_down_exps.weight iq4_nl +blk.7.ffn_gate_exps.weight iq4_nl +blk.7.ffn_up_exps.weight iq4_nl +blk.8.attn_k.weight iq4_nl +blk.8.attn_q.weight iq4_nl +blk.8.attn_v.weight q5_0 +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_gate_exps.weight iq4_nl +blk.8.ffn_up_exps.weight iq4_nl +blk.9.attn_k.weight iq4_nl +blk.9.attn_q.weight iq4_nl +blk.9.attn_v.weight q5_0 +blk.9.ffn_down_exps.weight iq4_nl +blk.9.ffn_gate_exps.weight iq4_nl +blk.9.ffn_up_exps.weight iq4_nl +blk.10.attn_k.weight iq4_nl +blk.10.attn_q.weight iq4_nl +blk.10.attn_v.weight q5_0 +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_gate_exps.weight iq4_nl +blk.10.ffn_up_exps.weight iq4_nl +blk.11.attn_k.weight iq4_nl +blk.11.attn_q.weight iq4_nl +blk.11.attn_v.weight q5_0 +blk.11.ffn_down_exps.weight iq4_nl +blk.11.ffn_gate_exps.weight iq4_nl +blk.11.ffn_up_exps.weight iq4_nl +blk.12.attn_k.weight iq4_nl +blk.12.attn_q.weight iq4_nl +blk.12.attn_v.weight q5_0 +blk.12.ffn_down_exps.weight iq4_nl +blk.12.ffn_gate_exps.weight iq4_nl +blk.12.ffn_up_exps.weight iq4_nl +blk.13.attn_k.weight iq4_nl +blk.13.attn_q.weight iq4_nl +blk.13.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_gate_exps.weight iq4_nl +blk.13.ffn_up_exps.weight iq4_nl +blk.14.attn_k.weight iq4_nl +blk.14.attn_q.weight iq4_nl +blk.14.attn_v.weight q5_0 +blk.14.ffn_down_exps.weight iq4_nl +blk.14.ffn_gate_exps.weight iq4_nl +blk.14.ffn_up_exps.weight iq4_nl +blk.15.attn_k.weight iq4_nl +blk.15.attn_q.weight iq4_nl +blk.15.attn_v.weight q5_0 +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_gate_exps.weight iq4_nl +blk.15.ffn_up_exps.weight iq4_nl +blk.16.attn_k.weight iq4_nl +blk.16.attn_q.weight iq4_nl +blk.16.attn_v.weight q5_0 +blk.16.ffn_down_exps.weight iq4_nl +blk.16.ffn_gate_exps.weight iq4_nl +blk.16.ffn_up_exps.weight iq4_nl +blk.17.attn_k.weight iq4_nl +blk.17.attn_q.weight iq4_nl +blk.17.attn_v.weight q5_0 +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_gate_exps.weight iq4_nl +blk.17.ffn_up_exps.weight iq4_nl +blk.18.attn_k.weight iq4_nl +blk.18.attn_q.weight iq4_nl +blk.18.attn_v.weight q5_0 +blk.18.ffn_down_exps.weight iq4_nl +blk.18.ffn_gate_exps.weight iq4_nl +blk.18.ffn_up_exps.weight iq4_nl +blk.19.attn_k.weight iq4_nl +blk.19.attn_q.weight iq4_nl +blk.19.attn_v.weight q5_0 +blk.19.ffn_down_exps.weight iq4_nl +blk.19.ffn_gate_exps.weight iq4_nl +blk.19.ffn_up_exps.weight iq4_nl +blk.20.attn_k.weight iq4_nl +blk.20.attn_q.weight iq4_nl +blk.20.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_gate_exps.weight iq4_nl +blk.20.ffn_up_exps.weight iq4_nl +blk.21.attn_k.weight iq4_nl +blk.21.attn_q.weight iq4_nl +blk.21.attn_v.weight q5_0 +blk.21.ffn_down_exps.weight iq4_nl +blk.21.ffn_gate_exps.weight iq4_nl +blk.21.ffn_up_exps.weight iq4_nl +blk.22.attn_k.weight iq4_nl +blk.22.attn_q.weight iq4_nl +blk.22.attn_v.weight q5_0 +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_gate_exps.weight iq4_nl +blk.22.ffn_up_exps.weight iq4_nl +blk.23.attn_k.weight iq4_nl +blk.23.attn_q.weight iq4_nl +blk.23.attn_v.weight q5_0 +blk.23.ffn_down_exps.weight iq4_nl +blk.23.ffn_gate_exps.weight iq4_nl +blk.23.ffn_up_exps.weight iq4_nl +blk.24.attn_k.weight iq4_nl +blk.24.attn_q.weight iq4_nl +blk.24.attn_v.weight q5_0 +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_gate_exps.weight iq4_nl +blk.24.ffn_up_exps.weight iq4_nl +blk.25.attn_k.weight iq4_nl +blk.25.attn_q.weight iq4_nl +blk.25.attn_v.weight q5_0 +blk.25.ffn_down_exps.weight iq4_nl +blk.25.ffn_gate_exps.weight iq4_nl +blk.25.ffn_up_exps.weight iq4_nl +blk.26.attn_k.weight iq4_nl +blk.26.attn_q.weight iq4_nl +blk.26.attn_v.weight q5_0 +blk.26.ffn_down_exps.weight iq4_nl +blk.26.ffn_gate_exps.weight iq4_nl +blk.26.ffn_up_exps.weight iq4_nl +blk.27.attn_k.weight iq4_nl +blk.27.attn_q.weight iq4_nl +blk.27.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_gate_exps.weight iq4_nl +blk.27.ffn_up_exps.weight iq4_nl +blk.28.attn_k.weight iq4_nl +blk.28.attn_q.weight iq4_nl +blk.28.attn_v.weight q5_0 +blk.28.ffn_down_exps.weight iq4_nl +blk.28.ffn_gate_exps.weight iq4_nl +blk.28.ffn_up_exps.weight iq4_nl +blk.29.attn_k.weight iq4_nl +blk.29.attn_q.weight iq4_nl +blk.29.attn_v.weight q5_0 +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_gate_exps.weight iq4_nl +blk.29.ffn_up_exps.weight iq4_nl +blk.30.attn_k.weight iq4_nl +blk.30.attn_q.weight iq4_nl +blk.30.attn_v.weight q5_0 +blk.30.ffn_down_exps.weight iq4_nl +blk.30.ffn_gate_exps.weight iq4_nl +blk.30.ffn_up_exps.weight iq4_nl +blk.31.attn_k.weight iq4_nl +blk.31.attn_q.weight iq4_nl +blk.31.attn_v.weight q5_0 +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_gate_exps.weight iq4_nl +blk.31.ffn_up_exps.weight iq4_nl +blk.32.attn_k.weight iq4_nl +blk.32.attn_q.weight iq4_nl +blk.32.attn_v.weight q5_0 +blk.32.ffn_down_exps.weight iq4_nl +blk.32.ffn_gate_exps.weight iq4_nl +blk.32.ffn_up_exps.weight iq4_nl +blk.33.attn_k.weight iq4_nl +blk.33.attn_q.weight iq4_nl +blk.33.attn_v.weight q5_0 +blk.33.ffn_down_exps.weight iq4_nl +blk.33.ffn_gate_exps.weight iq4_nl +blk.33.ffn_up_exps.weight iq4_nl +blk.34.attn_k.weight iq4_nl +blk.34.attn_q.weight iq4_nl +blk.34.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_gate_exps.weight iq4_nl +blk.34.ffn_up_exps.weight iq4_nl +blk.35.attn_k.weight iq4_nl +blk.35.attn_q.weight iq4_nl +blk.35.attn_v.weight q5_0 +blk.35.ffn_down_exps.weight iq4_nl +blk.35.ffn_gate_exps.weight iq4_nl +blk.35.ffn_up_exps.weight iq4_nl [IQ3_M] iq3_s output.weight q8_0 +token_embd.weight iq4_nl +blk.0.attn_k.weight iq4_nl blk.0.attn_output.weight q4_K -blk.0.attn_v.weight q4_K -blk.0.ffn_down_exps.weight q4_K +blk.0.attn_q.weight iq4_nl +blk.0.attn_v.weight q5_0 +blk.0.ffn_down_exps.weight q5_0 +blk.0.ffn_gate_exps.weight iq4_nl +blk.0.ffn_up_exps.weight iq4_nl +blk.1.attn_k.weight iq4_nl blk.1.attn_output.weight q4_K -blk.1.attn_v.weight q4_K -blk.1.ffn_down_exps.weight q4_K +blk.1.attn_q.weight iq4_nl +blk.1.attn_v.weight q5_0 +blk.1.ffn_down_exps.weight q5_0 +blk.1.ffn_gate_exps.weight iq4_nl +blk.1.ffn_up_exps.weight iq4_nl +blk.2.attn_k.weight iq4_nl blk.2.attn_output.weight q4_K -blk.2.attn_v.weight q4_K -blk.2.ffn_down_exps.weight q4_K +blk.2.attn_q.weight iq4_nl +blk.2.attn_v.weight q5_0 +blk.2.ffn_down_exps.weight q5_0 +blk.2.ffn_gate_exps.weight iq4_nl +blk.2.ffn_up_exps.weight iq4_nl +blk.3.attn_k.weight iq4_nl blk.3.attn_output.weight q4_K -blk.3.attn_v.weight q4_K -blk.3.ffn_down_exps.weight q4_K +blk.3.attn_q.weight iq4_nl +blk.3.attn_v.weight q5_0 +blk.3.ffn_down_exps.weight q5_0 +blk.3.ffn_gate_exps.weight iq4_nl +blk.3.ffn_up_exps.weight iq4_nl +blk.4.attn_k.weight iq4_nl blk.4.attn_output.weight q4_K -blk.4.attn_v.weight q4_K +blk.4.attn_q.weight iq4_nl +blk.4.attn_v.weight q5_0 +blk.4.ffn_down_exps.weight iq4_nl +blk.4.ffn_gate_exps.weight iq4_nl +blk.4.ffn_up_exps.weight iq4_nl +blk.5.attn_k.weight iq4_nl blk.5.attn_output.weight q4_K -blk.5.attn_v.weight q4_K +blk.5.attn_q.weight iq4_nl +blk.5.attn_v.weight q5_0 +blk.5.ffn_down_exps.weight iq4_nl +blk.5.ffn_gate_exps.weight iq4_nl +blk.5.ffn_up_exps.weight iq4_nl +blk.6.attn_k.weight iq4_nl blk.6.attn_output.weight q4_K -blk.6.attn_v.weight q4_K +blk.6.attn_q.weight iq4_nl +blk.6.attn_v.weight q5_0 +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_gate_exps.weight iq4_nl +blk.6.ffn_up_exps.weight iq4_nl +blk.7.attn_k.weight iq4_nl blk.7.attn_output.weight q4_K -blk.7.attn_v.weight q4_K +blk.7.attn_q.weight iq4_nl +blk.7.attn_v.weight q5_0 +blk.7.ffn_down_exps.weight iq4_nl +blk.7.ffn_gate_exps.weight iq4_nl +blk.7.ffn_up_exps.weight iq4_nl +blk.8.attn_k.weight iq4_nl blk.8.attn_output.weight q4_K -blk.8.attn_v.weight q4_K +blk.8.attn_q.weight iq4_nl +blk.8.attn_v.weight q5_0 +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_gate_exps.weight iq4_nl +blk.8.ffn_up_exps.weight iq4_nl +blk.9.attn_k.weight iq4_nl blk.9.attn_output.weight q4_K -blk.9.attn_v.weight q4_K +blk.9.attn_q.weight iq4_nl +blk.9.attn_v.weight q5_0 +blk.9.ffn_down_exps.weight iq4_nl +blk.9.ffn_gate_exps.weight iq4_nl +blk.9.ffn_up_exps.weight iq4_nl +blk.10.attn_k.weight iq4_nl blk.10.attn_output.weight q4_K -blk.10.attn_v.weight q4_K +blk.10.attn_q.weight iq4_nl +blk.10.attn_v.weight q5_0 +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_gate_exps.weight iq4_nl +blk.10.ffn_up_exps.weight iq4_nl +blk.11.attn_k.weight iq4_nl blk.11.attn_output.weight q4_K -blk.11.attn_v.weight q4_K +blk.11.attn_q.weight iq4_nl +blk.11.attn_v.weight q5_0 +blk.11.ffn_down_exps.weight iq4_nl +blk.11.ffn_gate_exps.weight iq4_nl +blk.11.ffn_up_exps.weight iq4_nl +blk.12.attn_k.weight iq4_nl blk.12.attn_output.weight q4_K -blk.12.attn_v.weight q4_K +blk.12.attn_q.weight iq4_nl +blk.12.attn_v.weight q5_0 +blk.12.ffn_down_exps.weight iq4_nl +blk.12.ffn_gate_exps.weight iq4_nl +blk.12.ffn_up_exps.weight iq4_nl +blk.13.attn_k.weight iq4_nl blk.13.attn_output.weight q4_K -blk.13.attn_v.weight q4_K +blk.13.attn_q.weight iq4_nl +blk.13.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_gate_exps.weight iq4_nl +blk.13.ffn_up_exps.weight iq4_nl +blk.14.attn_k.weight iq4_nl blk.14.attn_output.weight q4_K -blk.14.attn_v.weight q4_K +blk.14.attn_q.weight iq4_nl +blk.14.attn_v.weight q5_0 +blk.14.ffn_down_exps.weight iq4_nl +blk.14.ffn_gate_exps.weight iq4_nl +blk.14.ffn_up_exps.weight iq4_nl +blk.15.attn_k.weight iq4_nl blk.15.attn_output.weight q4_K -blk.15.attn_v.weight q4_K +blk.15.attn_q.weight iq4_nl +blk.15.attn_v.weight q5_0 +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_gate_exps.weight iq4_nl +blk.15.ffn_up_exps.weight iq4_nl +blk.16.attn_k.weight iq4_nl blk.16.attn_output.weight q4_K -blk.16.attn_v.weight q4_K +blk.16.attn_q.weight iq4_nl +blk.16.attn_v.weight q5_0 +blk.16.ffn_down_exps.weight iq4_nl +blk.16.ffn_gate_exps.weight iq4_nl +blk.16.ffn_up_exps.weight iq4_nl +blk.17.attn_k.weight iq4_nl blk.17.attn_output.weight q4_K -blk.17.attn_v.weight q4_K +blk.17.attn_q.weight iq4_nl +blk.17.attn_v.weight q5_0 +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_gate_exps.weight iq4_nl +blk.17.ffn_up_exps.weight iq4_nl +blk.18.attn_k.weight iq4_nl blk.18.attn_output.weight q4_K -blk.18.attn_v.weight q4_K +blk.18.attn_q.weight iq4_nl +blk.18.attn_v.weight q5_0 +blk.18.ffn_down_exps.weight iq4_nl +blk.18.ffn_gate_exps.weight iq4_nl +blk.18.ffn_up_exps.weight iq4_nl +blk.19.attn_k.weight iq4_nl blk.19.attn_output.weight q4_K -blk.19.attn_v.weight q4_K +blk.19.attn_q.weight iq4_nl +blk.19.attn_v.weight q5_0 +blk.19.ffn_down_exps.weight iq4_nl +blk.19.ffn_gate_exps.weight iq4_nl +blk.19.ffn_up_exps.weight iq4_nl +blk.20.attn_k.weight iq4_nl blk.20.attn_output.weight q4_K -blk.20.attn_v.weight q4_K +blk.20.attn_q.weight iq4_nl +blk.20.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_gate_exps.weight iq4_nl +blk.20.ffn_up_exps.weight iq4_nl +blk.21.attn_k.weight iq4_nl blk.21.attn_output.weight q4_K -blk.21.attn_v.weight q4_K +blk.21.attn_q.weight iq4_nl +blk.21.attn_v.weight q5_0 +blk.21.ffn_down_exps.weight iq4_nl +blk.21.ffn_gate_exps.weight iq4_nl +blk.21.ffn_up_exps.weight iq4_nl +blk.22.attn_k.weight iq4_nl blk.22.attn_output.weight q4_K -blk.22.attn_v.weight q4_K +blk.22.attn_q.weight iq4_nl +blk.22.attn_v.weight q5_0 +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_gate_exps.weight iq4_nl +blk.22.ffn_up_exps.weight iq4_nl +blk.23.attn_k.weight iq4_nl blk.23.attn_output.weight q4_K -blk.23.attn_v.weight q4_K +blk.23.attn_q.weight iq4_nl +blk.23.attn_v.weight q5_0 +blk.23.ffn_down_exps.weight iq4_nl +blk.23.ffn_gate_exps.weight iq4_nl +blk.23.ffn_up_exps.weight iq4_nl +blk.24.attn_k.weight iq4_nl blk.24.attn_output.weight q4_K -blk.24.attn_v.weight q4_K +blk.24.attn_q.weight iq4_nl +blk.24.attn_v.weight q5_0 +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_gate_exps.weight iq4_nl +blk.24.ffn_up_exps.weight iq4_nl +blk.25.attn_k.weight iq4_nl blk.25.attn_output.weight q4_K -blk.25.attn_v.weight q4_K +blk.25.attn_q.weight iq4_nl +blk.25.attn_v.weight q5_0 +blk.25.ffn_down_exps.weight iq4_nl +blk.25.ffn_gate_exps.weight iq4_nl +blk.25.ffn_up_exps.weight iq4_nl +blk.26.attn_k.weight iq4_nl blk.26.attn_output.weight q4_K -blk.26.attn_v.weight q4_K +blk.26.attn_q.weight iq4_nl +blk.26.attn_v.weight q5_0 +blk.26.ffn_down_exps.weight iq4_nl +blk.26.ffn_gate_exps.weight iq4_nl +blk.26.ffn_up_exps.weight iq4_nl +blk.27.attn_k.weight iq4_nl blk.27.attn_output.weight q4_K -blk.27.attn_v.weight q4_K +blk.27.attn_q.weight iq4_nl +blk.27.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_gate_exps.weight iq4_nl +blk.27.ffn_up_exps.weight iq4_nl +blk.28.attn_k.weight iq4_nl blk.28.attn_output.weight q4_K -blk.28.attn_v.weight q4_K +blk.28.attn_q.weight iq4_nl +blk.28.attn_v.weight q5_0 +blk.28.ffn_down_exps.weight iq4_nl +blk.28.ffn_gate_exps.weight iq4_nl +blk.28.ffn_up_exps.weight iq4_nl +blk.29.attn_k.weight iq4_nl blk.29.attn_output.weight q4_K -blk.29.attn_v.weight q4_K +blk.29.attn_q.weight iq4_nl +blk.29.attn_v.weight q5_0 +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_gate_exps.weight iq4_nl +blk.29.ffn_up_exps.weight iq4_nl +blk.30.attn_k.weight iq4_nl blk.30.attn_output.weight q4_K -blk.30.attn_v.weight q4_K +blk.30.attn_q.weight iq4_nl +blk.30.attn_v.weight q5_0 +blk.30.ffn_down_exps.weight iq4_nl +blk.30.ffn_gate_exps.weight iq4_nl +blk.30.ffn_up_exps.weight iq4_nl +blk.31.attn_k.weight iq4_nl blk.31.attn_output.weight q4_K -blk.31.attn_v.weight q4_K +blk.31.attn_q.weight iq4_nl +blk.31.attn_v.weight q5_0 +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_gate_exps.weight iq4_nl +blk.31.ffn_up_exps.weight iq4_nl +blk.32.attn_k.weight iq4_nl blk.32.attn_output.weight q4_K -blk.32.attn_v.weight q4_K +blk.32.attn_q.weight iq4_nl +blk.32.attn_v.weight q5_0 +blk.32.ffn_down_exps.weight iq4_nl +blk.32.ffn_gate_exps.weight iq4_nl +blk.32.ffn_up_exps.weight iq4_nl +blk.33.attn_k.weight iq4_nl blk.33.attn_output.weight q4_K -blk.33.attn_v.weight q4_K +blk.33.attn_q.weight iq4_nl +blk.33.attn_v.weight q5_0 +blk.33.ffn_down_exps.weight iq4_nl +blk.33.ffn_gate_exps.weight iq4_nl +blk.33.ffn_up_exps.weight iq4_nl +blk.34.attn_k.weight iq4_nl blk.34.attn_output.weight q4_K -blk.34.attn_v.weight q4_K +blk.34.attn_q.weight iq4_nl +blk.34.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_gate_exps.weight iq4_nl +blk.34.ffn_up_exps.weight iq4_nl +blk.35.attn_k.weight iq4_nl blk.35.attn_output.weight q4_K -blk.35.attn_v.weight q4_K +blk.35.attn_q.weight iq4_nl +blk.35.attn_v.weight q5_0 +blk.35.ffn_down_exps.weight iq4_nl +blk.35.ffn_gate_exps.weight iq4_nl +blk.35.ffn_up_exps.weight iq4_nl [IQ2_S] iq2_xs output.weight q8_0 -token_embd.weight iq3_s +token_embd.weight iq4_nl +blk.0.attn_k.weight iq4_nl blk.0.attn_output.weight iq3_s -blk.0.attn_v.weight q4_K -blk.0.ffn_down_exps.weight iq3_s +blk.0.attn_q.weight iq4_nl +blk.0.attn_v.weight q5_0 +blk.0.ffn_down_exps.weight iq4_nl +blk.0.ffn_gate_exps.weight iq4_nl +blk.0.ffn_up_exps.weight iq4_nl +blk.1.attn_k.weight iq4_nl blk.1.attn_output.weight iq3_s -blk.1.attn_v.weight q4_K -blk.1.ffn_down_exps.weight iq3_s +blk.1.attn_q.weight iq4_nl +blk.1.attn_v.weight q5_0 +blk.1.ffn_down_exps.weight iq4_nl +blk.1.ffn_gate_exps.weight iq4_nl +blk.1.ffn_up_exps.weight iq4_nl +blk.2.attn_k.weight iq4_nl blk.2.attn_output.weight iq3_s -blk.2.attn_v.weight q4_K -blk.2.ffn_down_exps.weight iq3_s +blk.2.attn_q.weight iq4_nl +blk.2.attn_v.weight q5_0 +blk.2.ffn_down_exps.weight iq4_nl +blk.2.ffn_gate_exps.weight iq4_nl +blk.2.ffn_up_exps.weight iq4_nl +blk.3.attn_k.weight iq4_nl blk.3.attn_output.weight iq3_s -blk.3.attn_v.weight q4_K -blk.3.ffn_down_exps.weight iq3_s +blk.3.attn_q.weight iq4_nl +blk.3.attn_v.weight q5_0 +blk.3.ffn_down_exps.weight iq4_nl +blk.3.ffn_gate_exps.weight iq4_nl +blk.3.ffn_up_exps.weight iq4_nl +blk.4.attn_k.weight iq4_nl blk.4.attn_output.weight iq3_s -blk.4.attn_v.weight q4_K +blk.4.attn_q.weight iq4_nl +blk.4.attn_v.weight q5_0 +blk.4.ffn_down_exps.weight iq4_nl +blk.4.ffn_gate_exps.weight iq4_nl +blk.4.ffn_up_exps.weight iq4_nl +blk.5.attn_k.weight iq4_nl blk.5.attn_output.weight iq3_s -blk.5.attn_v.weight q4_K +blk.5.attn_q.weight iq4_nl +blk.5.attn_v.weight q5_0 +blk.5.ffn_down_exps.weight iq4_nl +blk.5.ffn_gate_exps.weight iq4_nl +blk.5.ffn_up_exps.weight iq4_nl +blk.6.attn_k.weight iq4_nl blk.6.attn_output.weight iq3_s -blk.6.attn_v.weight q4_K +blk.6.attn_q.weight iq4_nl +blk.6.attn_v.weight q5_0 +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_gate_exps.weight iq4_nl +blk.6.ffn_up_exps.weight iq4_nl +blk.7.attn_k.weight iq4_nl blk.7.attn_output.weight iq3_s -blk.7.attn_v.weight q4_K +blk.7.attn_q.weight iq4_nl +blk.7.attn_v.weight q5_0 +blk.7.ffn_down_exps.weight iq4_nl +blk.7.ffn_gate_exps.weight iq4_nl +blk.7.ffn_up_exps.weight iq4_nl +blk.8.attn_k.weight iq4_nl blk.8.attn_output.weight iq3_s -blk.8.attn_v.weight q4_K +blk.8.attn_q.weight iq4_nl +blk.8.attn_v.weight q5_0 +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_gate_exps.weight iq4_nl +blk.8.ffn_up_exps.weight iq4_nl +blk.9.attn_k.weight iq4_nl blk.9.attn_output.weight iq3_s -blk.9.attn_v.weight q4_K +blk.9.attn_q.weight iq4_nl +blk.9.attn_v.weight q5_0 +blk.9.ffn_down_exps.weight iq4_nl +blk.9.ffn_gate_exps.weight iq4_nl +blk.9.ffn_up_exps.weight iq4_nl +blk.10.attn_k.weight iq4_nl blk.10.attn_output.weight iq3_s -blk.10.attn_v.weight q4_K +blk.10.attn_q.weight iq4_nl +blk.10.attn_v.weight q5_0 +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_gate_exps.weight iq4_nl +blk.10.ffn_up_exps.weight iq4_nl +blk.11.attn_k.weight iq4_nl blk.11.attn_output.weight iq3_s -blk.11.attn_v.weight q4_K +blk.11.attn_q.weight iq4_nl +blk.11.attn_v.weight q5_0 +blk.11.ffn_down_exps.weight iq4_nl +blk.11.ffn_gate_exps.weight iq4_nl +blk.11.ffn_up_exps.weight iq4_nl +blk.12.attn_k.weight iq4_nl blk.12.attn_output.weight iq3_s -blk.12.attn_v.weight q4_K +blk.12.attn_q.weight iq4_nl +blk.12.attn_v.weight q5_0 +blk.12.ffn_down_exps.weight iq4_nl +blk.12.ffn_gate_exps.weight iq4_nl +blk.12.ffn_up_exps.weight iq4_nl +blk.13.attn_k.weight iq4_nl blk.13.attn_output.weight iq3_s -blk.13.attn_v.weight q4_K +blk.13.attn_q.weight iq4_nl +blk.13.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_gate_exps.weight iq4_nl +blk.13.ffn_up_exps.weight iq4_nl +blk.14.attn_k.weight iq4_nl blk.14.attn_output.weight iq3_s -blk.14.attn_v.weight q4_K +blk.14.attn_q.weight iq4_nl +blk.14.attn_v.weight q5_0 +blk.14.ffn_down_exps.weight iq4_nl +blk.14.ffn_gate_exps.weight iq4_nl +blk.14.ffn_up_exps.weight iq4_nl +blk.15.attn_k.weight iq4_nl blk.15.attn_output.weight iq3_s -blk.15.attn_v.weight q4_K +blk.15.attn_q.weight iq4_nl +blk.15.attn_v.weight q5_0 +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_gate_exps.weight iq4_nl +blk.15.ffn_up_exps.weight iq4_nl +blk.16.attn_k.weight iq4_nl blk.16.attn_output.weight iq3_s -blk.16.attn_v.weight q4_K +blk.16.attn_q.weight iq4_nl +blk.16.attn_v.weight q5_0 +blk.16.ffn_down_exps.weight iq4_nl +blk.16.ffn_gate_exps.weight iq4_nl +blk.16.ffn_up_exps.weight iq4_nl +blk.17.attn_k.weight iq4_nl blk.17.attn_output.weight iq3_s -blk.17.attn_v.weight q4_K +blk.17.attn_q.weight iq4_nl +blk.17.attn_v.weight q5_0 +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_gate_exps.weight iq4_nl +blk.17.ffn_up_exps.weight iq4_nl +blk.18.attn_k.weight iq4_nl blk.18.attn_output.weight iq3_s -blk.18.attn_v.weight q4_K +blk.18.attn_q.weight iq4_nl +blk.18.attn_v.weight q5_0 +blk.18.ffn_down_exps.weight iq4_nl +blk.18.ffn_gate_exps.weight iq4_nl +blk.18.ffn_up_exps.weight iq4_nl +blk.19.attn_k.weight iq4_nl blk.19.attn_output.weight iq3_s -blk.19.attn_v.weight q4_K +blk.19.attn_q.weight iq4_nl +blk.19.attn_v.weight q5_0 +blk.19.ffn_down_exps.weight iq4_nl +blk.19.ffn_gate_exps.weight iq4_nl +blk.19.ffn_up_exps.weight iq4_nl +blk.20.attn_k.weight iq4_nl blk.20.attn_output.weight iq3_s -blk.20.attn_v.weight q4_K +blk.20.attn_q.weight iq4_nl +blk.20.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_gate_exps.weight iq4_nl +blk.20.ffn_up_exps.weight iq4_nl +blk.21.attn_k.weight iq4_nl blk.21.attn_output.weight iq3_s -blk.21.attn_v.weight q4_K +blk.21.attn_q.weight iq4_nl +blk.21.attn_v.weight q5_0 +blk.21.ffn_down_exps.weight iq4_nl +blk.21.ffn_gate_exps.weight iq4_nl +blk.21.ffn_up_exps.weight iq4_nl +blk.22.attn_k.weight iq4_nl blk.22.attn_output.weight iq3_s -blk.22.attn_v.weight q4_K +blk.22.attn_q.weight iq4_nl +blk.22.attn_v.weight q5_0 +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_gate_exps.weight iq4_nl +blk.22.ffn_up_exps.weight iq4_nl +blk.23.attn_k.weight iq4_nl blk.23.attn_output.weight iq3_s -blk.23.attn_v.weight q4_K +blk.23.attn_q.weight iq4_nl +blk.23.attn_v.weight q5_0 +blk.23.ffn_down_exps.weight iq4_nl +blk.23.ffn_gate_exps.weight iq4_nl +blk.23.ffn_up_exps.weight iq4_nl +blk.24.attn_k.weight iq4_nl blk.24.attn_output.weight iq3_s -blk.24.attn_v.weight q4_K +blk.24.attn_q.weight iq4_nl +blk.24.attn_v.weight q5_0 +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_gate_exps.weight iq4_nl +blk.24.ffn_up_exps.weight iq4_nl +blk.25.attn_k.weight iq4_nl blk.25.attn_output.weight iq3_s -blk.25.attn_v.weight q4_K +blk.25.attn_q.weight iq4_nl +blk.25.attn_v.weight q5_0 +blk.25.ffn_down_exps.weight iq4_nl +blk.25.ffn_gate_exps.weight iq4_nl +blk.25.ffn_up_exps.weight iq4_nl +blk.26.attn_k.weight iq4_nl blk.26.attn_output.weight iq3_s -blk.26.attn_v.weight q4_K +blk.26.attn_q.weight iq4_nl +blk.26.attn_v.weight q5_0 +blk.26.ffn_down_exps.weight iq4_nl +blk.26.ffn_gate_exps.weight iq4_nl +blk.26.ffn_up_exps.weight iq4_nl +blk.27.attn_k.weight iq4_nl blk.27.attn_output.weight iq3_s -blk.27.attn_v.weight q4_K +blk.27.attn_q.weight iq4_nl +blk.27.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_gate_exps.weight iq4_nl +blk.27.ffn_up_exps.weight iq4_nl +blk.28.attn_k.weight iq4_nl blk.28.attn_output.weight iq3_s -blk.28.attn_v.weight q4_K +blk.28.attn_q.weight iq4_nl +blk.28.attn_v.weight q5_0 +blk.28.ffn_down_exps.weight iq4_nl +blk.28.ffn_gate_exps.weight iq4_nl +blk.28.ffn_up_exps.weight iq4_nl +blk.29.attn_k.weight iq4_nl blk.29.attn_output.weight iq3_s -blk.29.attn_v.weight q4_K +blk.29.attn_q.weight iq4_nl +blk.29.attn_v.weight q5_0 +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_gate_exps.weight iq4_nl +blk.29.ffn_up_exps.weight iq4_nl +blk.30.attn_k.weight iq4_nl blk.30.attn_output.weight iq3_s -blk.30.attn_v.weight q4_K +blk.30.attn_q.weight iq4_nl +blk.30.attn_v.weight q5_0 +blk.30.ffn_down_exps.weight iq4_nl +blk.30.ffn_gate_exps.weight iq4_nl +blk.30.ffn_up_exps.weight iq4_nl +blk.31.attn_k.weight iq4_nl blk.31.attn_output.weight iq3_s -blk.31.attn_v.weight q4_K +blk.31.attn_q.weight iq4_nl +blk.31.attn_v.weight q5_0 +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_gate_exps.weight iq4_nl +blk.31.ffn_up_exps.weight iq4_nl +blk.32.attn_k.weight iq4_nl blk.32.attn_output.weight iq3_s -blk.32.attn_v.weight q4_K +blk.32.attn_q.weight iq4_nl +blk.32.attn_v.weight q5_0 +blk.32.ffn_down_exps.weight iq4_nl +blk.32.ffn_gate_exps.weight iq4_nl +blk.32.ffn_up_exps.weight iq4_nl +blk.33.attn_k.weight iq4_nl blk.33.attn_output.weight iq3_s -blk.33.attn_v.weight q4_K +blk.33.attn_q.weight iq4_nl +blk.33.attn_v.weight q5_0 +blk.33.ffn_down_exps.weight iq4_nl +blk.33.ffn_gate_exps.weight iq4_nl +blk.33.ffn_up_exps.weight iq4_nl +blk.34.attn_k.weight iq4_nl blk.34.attn_output.weight iq3_s -blk.34.attn_v.weight q4_K +blk.34.attn_q.weight iq4_nl +blk.34.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_gate_exps.weight iq4_nl +blk.34.ffn_up_exps.weight iq4_nl +blk.35.attn_k.weight iq4_nl blk.35.attn_output.weight iq3_s -blk.35.attn_v.weight q4_K +blk.35.attn_q.weight iq4_nl +blk.35.attn_v.weight q5_0 +blk.35.ffn_down_exps.weight iq4_nl +blk.35.ffn_gate_exps.weight iq4_nl +blk.35.ffn_up_exps.weight iq4_nl [IQ2_M] iq2_s output.weight q8_0 -token_embd.weight iq3_s +token_embd.weight iq4_nl +blk.0.attn_k.weight iq4_nl blk.0.attn_output.weight iq3_s -blk.0.attn_v.weight q4_K -blk.0.ffn_down_exps.weight iq3_s +blk.0.attn_q.weight iq4_nl +blk.0.attn_v.weight q5_0 +blk.0.ffn_down_exps.weight iq4_nl +blk.0.ffn_gate_exps.weight iq4_nl +blk.0.ffn_up_exps.weight iq4_nl +blk.1.attn_k.weight iq4_nl blk.1.attn_output.weight iq3_s -blk.1.attn_v.weight q4_K -blk.1.ffn_down_exps.weight iq3_s +blk.1.attn_q.weight iq4_nl +blk.1.attn_v.weight q5_0 +blk.1.ffn_down_exps.weight iq4_nl +blk.1.ffn_gate_exps.weight iq4_nl +blk.1.ffn_up_exps.weight iq4_nl +blk.2.attn_k.weight iq4_nl blk.2.attn_output.weight iq3_s -blk.2.attn_v.weight q4_K -blk.2.ffn_down_exps.weight iq3_s +blk.2.attn_q.weight iq4_nl +blk.2.attn_v.weight q5_0 +blk.2.ffn_down_exps.weight iq4_nl +blk.2.ffn_gate_exps.weight iq4_nl +blk.2.ffn_up_exps.weight iq4_nl +blk.3.attn_k.weight iq4_nl blk.3.attn_output.weight iq3_s -blk.3.attn_v.weight q4_K -blk.3.ffn_down_exps.weight iq3_s +blk.3.attn_q.weight iq4_nl +blk.3.attn_v.weight q5_0 +blk.3.ffn_down_exps.weight iq4_nl +blk.3.ffn_gate_exps.weight iq4_nl +blk.3.ffn_up_exps.weight iq4_nl +blk.4.attn_k.weight iq4_nl blk.4.attn_output.weight iq3_s -blk.4.attn_v.weight q4_K +blk.4.attn_q.weight iq4_nl +blk.4.attn_v.weight q5_0 +blk.4.ffn_down_exps.weight iq4_nl +blk.4.ffn_gate_exps.weight iq4_nl +blk.4.ffn_up_exps.weight iq4_nl +blk.5.attn_k.weight iq4_nl blk.5.attn_output.weight iq3_s -blk.5.attn_v.weight q4_K +blk.5.attn_q.weight iq4_nl +blk.5.attn_v.weight q5_0 +blk.5.ffn_down_exps.weight iq4_nl +blk.5.ffn_gate_exps.weight iq4_nl +blk.5.ffn_up_exps.weight iq4_nl +blk.6.attn_k.weight iq4_nl blk.6.attn_output.weight iq3_s -blk.6.attn_v.weight q4_K +blk.6.attn_q.weight iq4_nl +blk.6.attn_v.weight q5_0 +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_gate_exps.weight iq4_nl +blk.6.ffn_up_exps.weight iq4_nl +blk.7.attn_k.weight iq4_nl blk.7.attn_output.weight iq3_s -blk.7.attn_v.weight q4_K +blk.7.attn_q.weight iq4_nl +blk.7.attn_v.weight q5_0 +blk.7.ffn_down_exps.weight iq4_nl +blk.7.ffn_gate_exps.weight iq4_nl +blk.7.ffn_up_exps.weight iq4_nl +blk.8.attn_k.weight iq4_nl blk.8.attn_output.weight iq3_s -blk.8.attn_v.weight q4_K +blk.8.attn_q.weight iq4_nl +blk.8.attn_v.weight q5_0 +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_gate_exps.weight iq4_nl +blk.8.ffn_up_exps.weight iq4_nl +blk.9.attn_k.weight iq4_nl blk.9.attn_output.weight iq3_s -blk.9.attn_v.weight q4_K +blk.9.attn_q.weight iq4_nl +blk.9.attn_v.weight q5_0 +blk.9.ffn_down_exps.weight iq4_nl +blk.9.ffn_gate_exps.weight iq4_nl +blk.9.ffn_up_exps.weight iq4_nl +blk.10.attn_k.weight iq4_nl blk.10.attn_output.weight iq3_s -blk.10.attn_v.weight q4_K +blk.10.attn_q.weight iq4_nl +blk.10.attn_v.weight q5_0 +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_gate_exps.weight iq4_nl +blk.10.ffn_up_exps.weight iq4_nl +blk.11.attn_k.weight iq4_nl blk.11.attn_output.weight iq3_s -blk.11.attn_v.weight q4_K +blk.11.attn_q.weight iq4_nl +blk.11.attn_v.weight q5_0 +blk.11.ffn_down_exps.weight iq4_nl +blk.11.ffn_gate_exps.weight iq4_nl +blk.11.ffn_up_exps.weight iq4_nl +blk.12.attn_k.weight iq4_nl blk.12.attn_output.weight iq3_s -blk.12.attn_v.weight q4_K +blk.12.attn_q.weight iq4_nl +blk.12.attn_v.weight q5_0 +blk.12.ffn_down_exps.weight iq4_nl +blk.12.ffn_gate_exps.weight iq4_nl +blk.12.ffn_up_exps.weight iq4_nl +blk.13.attn_k.weight iq4_nl blk.13.attn_output.weight iq3_s -blk.13.attn_v.weight q4_K +blk.13.attn_q.weight iq4_nl +blk.13.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_gate_exps.weight iq4_nl +blk.13.ffn_up_exps.weight iq4_nl +blk.14.attn_k.weight iq4_nl blk.14.attn_output.weight iq3_s -blk.14.attn_v.weight q4_K +blk.14.attn_q.weight iq4_nl +blk.14.attn_v.weight q5_0 +blk.14.ffn_down_exps.weight iq4_nl +blk.14.ffn_gate_exps.weight iq4_nl +blk.14.ffn_up_exps.weight iq4_nl +blk.15.attn_k.weight iq4_nl blk.15.attn_output.weight iq3_s -blk.15.attn_v.weight q4_K +blk.15.attn_q.weight iq4_nl +blk.15.attn_v.weight q5_0 +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_gate_exps.weight iq4_nl +blk.15.ffn_up_exps.weight iq4_nl +blk.16.attn_k.weight iq4_nl blk.16.attn_output.weight iq3_s -blk.16.attn_v.weight q4_K +blk.16.attn_q.weight iq4_nl +blk.16.attn_v.weight q5_0 +blk.16.ffn_down_exps.weight iq4_nl +blk.16.ffn_gate_exps.weight iq4_nl +blk.16.ffn_up_exps.weight iq4_nl +blk.17.attn_k.weight iq4_nl blk.17.attn_output.weight iq3_s -blk.17.attn_v.weight q4_K +blk.17.attn_q.weight iq4_nl +blk.17.attn_v.weight q5_0 +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_gate_exps.weight iq4_nl +blk.17.ffn_up_exps.weight iq4_nl +blk.18.attn_k.weight iq4_nl blk.18.attn_output.weight iq3_s -blk.18.attn_v.weight q4_K +blk.18.attn_q.weight iq4_nl +blk.18.attn_v.weight q5_0 +blk.18.ffn_down_exps.weight iq4_nl +blk.18.ffn_gate_exps.weight iq4_nl +blk.18.ffn_up_exps.weight iq4_nl +blk.19.attn_k.weight iq4_nl blk.19.attn_output.weight iq3_s -blk.19.attn_v.weight q4_K +blk.19.attn_q.weight iq4_nl +blk.19.attn_v.weight q5_0 +blk.19.ffn_down_exps.weight iq4_nl +blk.19.ffn_gate_exps.weight iq4_nl +blk.19.ffn_up_exps.weight iq4_nl +blk.20.attn_k.weight iq4_nl blk.20.attn_output.weight iq3_s -blk.20.attn_v.weight q4_K +blk.20.attn_q.weight iq4_nl +blk.20.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_gate_exps.weight iq4_nl +blk.20.ffn_up_exps.weight iq4_nl +blk.21.attn_k.weight iq4_nl blk.21.attn_output.weight iq3_s -blk.21.attn_v.weight q4_K +blk.21.attn_q.weight iq4_nl +blk.21.attn_v.weight q5_0 +blk.21.ffn_down_exps.weight iq4_nl +blk.21.ffn_gate_exps.weight iq4_nl +blk.21.ffn_up_exps.weight iq4_nl +blk.22.attn_k.weight iq4_nl blk.22.attn_output.weight iq3_s -blk.22.attn_v.weight q4_K +blk.22.attn_q.weight iq4_nl +blk.22.attn_v.weight q5_0 +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_gate_exps.weight iq4_nl +blk.22.ffn_up_exps.weight iq4_nl +blk.23.attn_k.weight iq4_nl blk.23.attn_output.weight iq3_s -blk.23.attn_v.weight q4_K +blk.23.attn_q.weight iq4_nl +blk.23.attn_v.weight q5_0 +blk.23.ffn_down_exps.weight iq4_nl +blk.23.ffn_gate_exps.weight iq4_nl +blk.23.ffn_up_exps.weight iq4_nl +blk.24.attn_k.weight iq4_nl blk.24.attn_output.weight iq3_s -blk.24.attn_v.weight q4_K +blk.24.attn_q.weight iq4_nl +blk.24.attn_v.weight q5_0 +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_gate_exps.weight iq4_nl +blk.24.ffn_up_exps.weight iq4_nl +blk.25.attn_k.weight iq4_nl blk.25.attn_output.weight iq3_s -blk.25.attn_v.weight q4_K +blk.25.attn_q.weight iq4_nl +blk.25.attn_v.weight q5_0 +blk.25.ffn_down_exps.weight iq4_nl +blk.25.ffn_gate_exps.weight iq4_nl +blk.25.ffn_up_exps.weight iq4_nl +blk.26.attn_k.weight iq4_nl blk.26.attn_output.weight iq3_s -blk.26.attn_v.weight q4_K +blk.26.attn_q.weight iq4_nl +blk.26.attn_v.weight q5_0 +blk.26.ffn_down_exps.weight iq4_nl +blk.26.ffn_gate_exps.weight iq4_nl +blk.26.ffn_up_exps.weight iq4_nl +blk.27.attn_k.weight iq4_nl blk.27.attn_output.weight iq3_s -blk.27.attn_v.weight q4_K +blk.27.attn_q.weight iq4_nl +blk.27.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_gate_exps.weight iq4_nl +blk.27.ffn_up_exps.weight iq4_nl +blk.28.attn_k.weight iq4_nl blk.28.attn_output.weight iq3_s -blk.28.attn_v.weight q4_K +blk.28.attn_q.weight iq4_nl +blk.28.attn_v.weight q5_0 +blk.28.ffn_down_exps.weight iq4_nl +blk.28.ffn_gate_exps.weight iq4_nl +blk.28.ffn_up_exps.weight iq4_nl +blk.29.attn_k.weight iq4_nl blk.29.attn_output.weight iq3_s -blk.29.attn_v.weight q4_K +blk.29.attn_q.weight iq4_nl +blk.29.attn_v.weight q5_0 +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_gate_exps.weight iq4_nl +blk.29.ffn_up_exps.weight iq4_nl +blk.30.attn_k.weight iq4_nl blk.30.attn_output.weight iq3_s -blk.30.attn_v.weight q4_K +blk.30.attn_q.weight iq4_nl +blk.30.attn_v.weight q5_0 +blk.30.ffn_down_exps.weight iq4_nl +blk.30.ffn_gate_exps.weight iq4_nl +blk.30.ffn_up_exps.weight iq4_nl +blk.31.attn_k.weight iq4_nl blk.31.attn_output.weight iq3_s -blk.31.attn_v.weight q4_K +blk.31.attn_q.weight iq4_nl +blk.31.attn_v.weight q5_0 +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_gate_exps.weight iq4_nl +blk.31.ffn_up_exps.weight iq4_nl +blk.32.attn_k.weight iq4_nl blk.32.attn_output.weight iq3_s -blk.32.attn_v.weight q4_K +blk.32.attn_q.weight iq4_nl +blk.32.attn_v.weight q5_0 +blk.32.ffn_down_exps.weight iq4_nl +blk.32.ffn_gate_exps.weight iq4_nl +blk.32.ffn_up_exps.weight iq4_nl +blk.33.attn_k.weight iq4_nl blk.33.attn_output.weight iq3_s -blk.33.attn_v.weight q4_K +blk.33.attn_q.weight iq4_nl +blk.33.attn_v.weight q5_0 +blk.33.ffn_down_exps.weight iq4_nl +blk.33.ffn_gate_exps.weight iq4_nl +blk.33.ffn_up_exps.weight iq4_nl +blk.34.attn_k.weight iq4_nl blk.34.attn_output.weight iq3_s -blk.34.attn_v.weight q4_K +blk.34.attn_q.weight iq4_nl +blk.34.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_gate_exps.weight iq4_nl +blk.34.ffn_up_exps.weight iq4_nl +blk.35.attn_k.weight iq4_nl blk.35.attn_output.weight iq3_s -blk.35.attn_v.weight q4_K +blk.35.attn_q.weight iq4_nl +blk.35.attn_v.weight q5_0 +blk.35.ffn_down_exps.weight iq4_nl +blk.35.ffn_gate_exps.weight iq4_nl +blk.35.ffn_up_exps.weight iq4_nl [IQ4_XS] iq4_xs output.weight q8_0 -blk.0.attn_v.weight q5_K -blk.0.ffn_down_exps.weight q5_K -blk.1.attn_v.weight q5_K -blk.1.ffn_down_exps.weight q5_K -blk.2.attn_v.weight q5_K -blk.2.ffn_down_exps.weight q5_K -blk.3.attn_v.weight q5_K -blk.3.ffn_down_exps.weight q5_K -blk.4.attn_v.weight q5_K -blk.5.attn_v.weight q5_K -blk.6.attn_v.weight q5_K -blk.7.attn_v.weight q5_K -blk.8.attn_v.weight q5_K -blk.9.attn_v.weight q5_K -blk.10.attn_v.weight q5_K -blk.11.attn_v.weight q5_K -blk.12.attn_v.weight q5_K -blk.13.attn_v.weight q5_K -blk.14.attn_v.weight q5_K -blk.15.attn_v.weight q5_K -blk.16.attn_v.weight q5_K -blk.17.attn_v.weight q5_K -blk.18.attn_v.weight q5_K -blk.19.attn_v.weight q5_K -blk.20.attn_v.weight q5_K -blk.21.attn_v.weight q5_K -blk.22.attn_v.weight q5_K -blk.23.attn_v.weight q5_K -blk.24.attn_v.weight q5_K -blk.25.attn_v.weight q5_K -blk.26.attn_v.weight q5_K -blk.27.attn_v.weight q5_K -blk.28.attn_v.weight q5_K -blk.29.attn_v.weight q5_K -blk.30.attn_v.weight q5_K -blk.31.attn_v.weight q5_K -blk.32.attn_v.weight q5_K -blk.33.attn_v.weight q5_K -blk.34.attn_v.weight q5_K -blk.35.attn_v.weight q5_K +token_embd.weight iq4_nl +blk.0.attn_k.weight iq4_nl +blk.0.attn_q.weight iq4_nl +blk.0.attn_v.weight q5_1 +blk.0.ffn_down_exps.weight q5_1 +blk.0.ffn_gate_exps.weight iq4_nl +blk.0.ffn_up_exps.weight iq4_nl +blk.1.attn_k.weight iq4_nl +blk.1.attn_q.weight iq4_nl +blk.1.attn_v.weight q5_1 +blk.1.ffn_down_exps.weight q5_1 +blk.1.ffn_gate_exps.weight iq4_nl +blk.1.ffn_up_exps.weight iq4_nl +blk.2.attn_k.weight iq4_nl +blk.2.attn_q.weight iq4_nl +blk.2.attn_v.weight q5_1 +blk.2.ffn_down_exps.weight q5_1 +blk.2.ffn_gate_exps.weight iq4_nl +blk.2.ffn_up_exps.weight iq4_nl +blk.3.attn_k.weight iq4_nl +blk.3.attn_q.weight iq4_nl +blk.3.attn_v.weight q5_1 +blk.3.ffn_down_exps.weight q5_1 +blk.3.ffn_gate_exps.weight iq4_nl +blk.3.ffn_up_exps.weight iq4_nl +blk.4.attn_k.weight iq4_nl +blk.4.attn_q.weight iq4_nl +blk.4.attn_v.weight q5_1 +blk.4.ffn_down_exps.weight iq4_nl +blk.4.ffn_gate_exps.weight iq4_nl +blk.4.ffn_up_exps.weight iq4_nl +blk.5.attn_k.weight iq4_nl +blk.5.attn_q.weight iq4_nl +blk.5.attn_v.weight q5_1 +blk.5.ffn_down_exps.weight iq4_nl +blk.5.ffn_gate_exps.weight iq4_nl +blk.5.ffn_up_exps.weight iq4_nl +blk.6.attn_k.weight iq4_nl +blk.6.attn_q.weight iq4_nl +blk.6.attn_v.weight q5_1 +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_gate_exps.weight iq4_nl +blk.6.ffn_up_exps.weight iq4_nl +blk.7.attn_k.weight iq4_nl +blk.7.attn_q.weight iq4_nl +blk.7.attn_v.weight q5_1 +blk.7.ffn_down_exps.weight iq4_nl +blk.7.ffn_gate_exps.weight iq4_nl +blk.7.ffn_up_exps.weight iq4_nl +blk.8.attn_k.weight iq4_nl +blk.8.attn_q.weight iq4_nl +blk.8.attn_v.weight q5_1 +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_gate_exps.weight iq4_nl +blk.8.ffn_up_exps.weight iq4_nl +blk.9.attn_k.weight iq4_nl +blk.9.attn_q.weight iq4_nl +blk.9.attn_v.weight q5_1 +blk.9.ffn_down_exps.weight iq4_nl +blk.9.ffn_gate_exps.weight iq4_nl +blk.9.ffn_up_exps.weight iq4_nl +blk.10.attn_k.weight iq4_nl +blk.10.attn_q.weight iq4_nl +blk.10.attn_v.weight q5_1 +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_gate_exps.weight iq4_nl +blk.10.ffn_up_exps.weight iq4_nl +blk.11.attn_k.weight iq4_nl +blk.11.attn_q.weight iq4_nl +blk.11.attn_v.weight q5_1 +blk.11.ffn_down_exps.weight iq4_nl +blk.11.ffn_gate_exps.weight iq4_nl +blk.11.ffn_up_exps.weight iq4_nl +blk.12.attn_k.weight iq4_nl +blk.12.attn_q.weight iq4_nl +blk.12.attn_v.weight q5_1 +blk.12.ffn_down_exps.weight iq4_nl +blk.12.ffn_gate_exps.weight iq4_nl +blk.12.ffn_up_exps.weight iq4_nl +blk.13.attn_k.weight iq4_nl +blk.13.attn_q.weight iq4_nl +blk.13.attn_v.weight q5_1 +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_gate_exps.weight iq4_nl +blk.13.ffn_up_exps.weight iq4_nl +blk.14.attn_k.weight iq4_nl +blk.14.attn_q.weight iq4_nl +blk.14.attn_v.weight q5_1 +blk.14.ffn_down_exps.weight iq4_nl +blk.14.ffn_gate_exps.weight iq4_nl +blk.14.ffn_up_exps.weight iq4_nl +blk.15.attn_k.weight iq4_nl +blk.15.attn_q.weight iq4_nl +blk.15.attn_v.weight q5_1 +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_gate_exps.weight iq4_nl +blk.15.ffn_up_exps.weight iq4_nl +blk.16.attn_k.weight iq4_nl +blk.16.attn_q.weight iq4_nl +blk.16.attn_v.weight q5_1 +blk.16.ffn_down_exps.weight iq4_nl +blk.16.ffn_gate_exps.weight iq4_nl +blk.16.ffn_up_exps.weight iq4_nl +blk.17.attn_k.weight iq4_nl +blk.17.attn_q.weight iq4_nl +blk.17.attn_v.weight q5_1 +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_gate_exps.weight iq4_nl +blk.17.ffn_up_exps.weight iq4_nl +blk.18.attn_k.weight iq4_nl +blk.18.attn_q.weight iq4_nl +blk.18.attn_v.weight q5_1 +blk.18.ffn_down_exps.weight iq4_nl +blk.18.ffn_gate_exps.weight iq4_nl +blk.18.ffn_up_exps.weight iq4_nl +blk.19.attn_k.weight iq4_nl +blk.19.attn_q.weight iq4_nl +blk.19.attn_v.weight q5_1 +blk.19.ffn_down_exps.weight iq4_nl +blk.19.ffn_gate_exps.weight iq4_nl +blk.19.ffn_up_exps.weight iq4_nl +blk.20.attn_k.weight iq4_nl +blk.20.attn_q.weight iq4_nl +blk.20.attn_v.weight q5_1 +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_gate_exps.weight iq4_nl +blk.20.ffn_up_exps.weight iq4_nl +blk.21.attn_k.weight iq4_nl +blk.21.attn_q.weight iq4_nl +blk.21.attn_v.weight q5_1 +blk.21.ffn_down_exps.weight iq4_nl +blk.21.ffn_gate_exps.weight iq4_nl +blk.21.ffn_up_exps.weight iq4_nl +blk.22.attn_k.weight iq4_nl +blk.22.attn_q.weight iq4_nl +blk.22.attn_v.weight q5_1 +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_gate_exps.weight iq4_nl +blk.22.ffn_up_exps.weight iq4_nl +blk.23.attn_k.weight iq4_nl +blk.23.attn_q.weight iq4_nl +blk.23.attn_v.weight q5_1 +blk.23.ffn_down_exps.weight iq4_nl +blk.23.ffn_gate_exps.weight iq4_nl +blk.23.ffn_up_exps.weight iq4_nl +blk.24.attn_k.weight iq4_nl +blk.24.attn_q.weight iq4_nl +blk.24.attn_v.weight q5_1 +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_gate_exps.weight iq4_nl +blk.24.ffn_up_exps.weight iq4_nl +blk.25.attn_k.weight iq4_nl +blk.25.attn_q.weight iq4_nl +blk.25.attn_v.weight q5_1 +blk.25.ffn_down_exps.weight iq4_nl +blk.25.ffn_gate_exps.weight iq4_nl +blk.25.ffn_up_exps.weight iq4_nl +blk.26.attn_k.weight iq4_nl +blk.26.attn_q.weight iq4_nl +blk.26.attn_v.weight q5_1 +blk.26.ffn_down_exps.weight iq4_nl +blk.26.ffn_gate_exps.weight iq4_nl +blk.26.ffn_up_exps.weight iq4_nl +blk.27.attn_k.weight iq4_nl +blk.27.attn_q.weight iq4_nl +blk.27.attn_v.weight q5_1 +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_gate_exps.weight iq4_nl +blk.27.ffn_up_exps.weight iq4_nl +blk.28.attn_k.weight iq4_nl +blk.28.attn_q.weight iq4_nl +blk.28.attn_v.weight q5_1 +blk.28.ffn_down_exps.weight iq4_nl +blk.28.ffn_gate_exps.weight iq4_nl +blk.28.ffn_up_exps.weight iq4_nl +blk.29.attn_k.weight iq4_nl +blk.29.attn_q.weight iq4_nl +blk.29.attn_v.weight q5_1 +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_gate_exps.weight iq4_nl +blk.29.ffn_up_exps.weight iq4_nl +blk.30.attn_k.weight iq4_nl +blk.30.attn_q.weight iq4_nl +blk.30.attn_v.weight q5_1 +blk.30.ffn_down_exps.weight iq4_nl +blk.30.ffn_gate_exps.weight iq4_nl +blk.30.ffn_up_exps.weight iq4_nl +blk.31.attn_k.weight iq4_nl +blk.31.attn_q.weight iq4_nl +blk.31.attn_v.weight q5_1 +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_gate_exps.weight iq4_nl +blk.31.ffn_up_exps.weight iq4_nl +blk.32.attn_k.weight iq4_nl +blk.32.attn_q.weight iq4_nl +blk.32.attn_v.weight q5_1 +blk.32.ffn_down_exps.weight iq4_nl +blk.32.ffn_gate_exps.weight iq4_nl +blk.32.ffn_up_exps.weight iq4_nl +blk.33.attn_k.weight iq4_nl +blk.33.attn_q.weight iq4_nl +blk.33.attn_v.weight q5_1 +blk.33.ffn_down_exps.weight iq4_nl +blk.33.ffn_gate_exps.weight iq4_nl +blk.33.ffn_up_exps.weight iq4_nl +blk.34.attn_k.weight iq4_nl +blk.34.attn_q.weight iq4_nl +blk.34.attn_v.weight q5_1 +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_gate_exps.weight iq4_nl +blk.34.ffn_up_exps.weight iq4_nl +blk.35.attn_k.weight iq4_nl +blk.35.attn_q.weight iq4_nl +blk.35.attn_v.weight q5_1 +blk.35.ffn_down_exps.weight iq4_nl +blk.35.ffn_gate_exps.weight iq4_nl +blk.35.ffn_up_exps.weight iq4_nl [IQ1_M] iq1_m output.weight q8_0 -token_embd.weight q2_K +token_embd.weight q4_0 +blk.0.attn_k.weight iq4_nl blk.0.attn_output.weight iq2_xxs -blk.0.attn_v.weight q4_K -blk.0.ffn_down_exps.weight q2_K +blk.0.attn_q.weight iq4_nl +blk.0.attn_v.weight q5_0 +blk.0.ffn_down_exps.weight q4_0 +blk.0.ffn_gate_exps.weight iq4_nl +blk.0.ffn_up_exps.weight iq4_nl +blk.1.attn_k.weight iq4_nl blk.1.attn_output.weight iq2_xxs -blk.1.attn_v.weight q4_K -blk.1.ffn_down_exps.weight q2_K +blk.1.attn_q.weight iq4_nl +blk.1.attn_v.weight q5_0 +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_gate_exps.weight iq4_nl +blk.1.ffn_up_exps.weight iq4_nl +blk.2.attn_k.weight iq4_nl blk.2.attn_output.weight iq2_xxs -blk.2.attn_v.weight q4_K -blk.2.ffn_down_exps.weight q2_K +blk.2.attn_q.weight iq4_nl +blk.2.attn_v.weight q5_0 +blk.2.ffn_down_exps.weight q4_0 +blk.2.ffn_gate_exps.weight iq4_nl +blk.2.ffn_up_exps.weight iq4_nl +blk.3.attn_k.weight iq4_nl blk.3.attn_output.weight iq2_xxs -blk.3.attn_v.weight q4_K -blk.3.ffn_down_exps.weight q2_K +blk.3.attn_q.weight iq4_nl +blk.3.attn_v.weight q5_0 +blk.3.ffn_down_exps.weight q4_0 +blk.3.ffn_gate_exps.weight iq4_nl +blk.3.ffn_up_exps.weight iq4_nl +blk.4.attn_k.weight iq4_nl blk.4.attn_output.weight iq2_xxs -blk.4.attn_v.weight q4_K +blk.4.attn_q.weight iq4_nl +blk.4.attn_v.weight q5_0 +blk.4.ffn_down_exps.weight iq4_nl +blk.4.ffn_gate_exps.weight iq4_nl +blk.4.ffn_up_exps.weight iq4_nl +blk.5.attn_k.weight iq4_nl blk.5.attn_output.weight iq2_xxs -blk.5.attn_v.weight q4_K +blk.5.attn_q.weight iq4_nl +blk.5.attn_v.weight q5_0 +blk.5.ffn_down_exps.weight iq4_nl +blk.5.ffn_gate_exps.weight iq4_nl +blk.5.ffn_up_exps.weight iq4_nl +blk.6.attn_k.weight iq4_nl blk.6.attn_output.weight iq2_xxs -blk.6.attn_v.weight q4_K +blk.6.attn_q.weight iq4_nl +blk.6.attn_v.weight q5_0 +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_gate_exps.weight iq4_nl +blk.6.ffn_up_exps.weight iq4_nl +blk.7.attn_k.weight iq4_nl blk.7.attn_output.weight iq2_xxs -blk.7.attn_v.weight q4_K +blk.7.attn_q.weight iq4_nl +blk.7.attn_v.weight q5_0 +blk.7.ffn_down_exps.weight iq4_nl +blk.7.ffn_gate_exps.weight iq4_nl +blk.7.ffn_up_exps.weight iq4_nl +blk.8.attn_k.weight iq4_nl blk.8.attn_output.weight iq2_xxs -blk.8.attn_v.weight q4_K +blk.8.attn_q.weight iq4_nl +blk.8.attn_v.weight q5_0 +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_gate_exps.weight iq4_nl +blk.8.ffn_up_exps.weight iq4_nl +blk.9.attn_k.weight iq4_nl blk.9.attn_output.weight iq2_xxs -blk.9.attn_v.weight q4_K +blk.9.attn_q.weight iq4_nl +blk.9.attn_v.weight q5_0 +blk.9.ffn_down_exps.weight iq4_nl +blk.9.ffn_gate_exps.weight iq4_nl +blk.9.ffn_up_exps.weight iq4_nl +blk.10.attn_k.weight iq4_nl blk.10.attn_output.weight iq2_xxs -blk.10.attn_v.weight q4_K +blk.10.attn_q.weight iq4_nl +blk.10.attn_v.weight q5_0 +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_gate_exps.weight iq4_nl +blk.10.ffn_up_exps.weight iq4_nl +blk.11.attn_k.weight iq4_nl blk.11.attn_output.weight iq2_xxs -blk.11.attn_v.weight q4_K +blk.11.attn_q.weight iq4_nl +blk.11.attn_v.weight q5_0 +blk.11.ffn_down_exps.weight iq4_nl +blk.11.ffn_gate_exps.weight iq4_nl +blk.11.ffn_up_exps.weight iq4_nl +blk.12.attn_k.weight iq4_nl blk.12.attn_output.weight iq2_xxs -blk.12.attn_v.weight q4_K +blk.12.attn_q.weight iq4_nl +blk.12.attn_v.weight q5_0 +blk.12.ffn_down_exps.weight iq4_nl +blk.12.ffn_gate_exps.weight iq4_nl +blk.12.ffn_up_exps.weight iq4_nl +blk.13.attn_k.weight iq4_nl blk.13.attn_output.weight iq2_xxs -blk.13.attn_v.weight q4_K +blk.13.attn_q.weight iq4_nl +blk.13.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_gate_exps.weight iq4_nl +blk.13.ffn_up_exps.weight iq4_nl +blk.14.attn_k.weight iq4_nl blk.14.attn_output.weight iq2_xxs -blk.14.attn_v.weight q4_K +blk.14.attn_q.weight iq4_nl +blk.14.attn_v.weight q5_0 +blk.14.ffn_down_exps.weight iq4_nl +blk.14.ffn_gate_exps.weight iq4_nl +blk.14.ffn_up_exps.weight iq4_nl +blk.15.attn_k.weight iq4_nl blk.15.attn_output.weight iq2_xxs -blk.15.attn_v.weight q4_K +blk.15.attn_q.weight iq4_nl +blk.15.attn_v.weight q5_0 +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_gate_exps.weight iq4_nl +blk.15.ffn_up_exps.weight iq4_nl +blk.16.attn_k.weight iq4_nl blk.16.attn_output.weight iq2_xxs -blk.16.attn_v.weight q4_K +blk.16.attn_q.weight iq4_nl +blk.16.attn_v.weight q5_0 +blk.16.ffn_down_exps.weight iq4_nl +blk.16.ffn_gate_exps.weight iq4_nl +blk.16.ffn_up_exps.weight iq4_nl +blk.17.attn_k.weight iq4_nl blk.17.attn_output.weight iq2_xxs -blk.17.attn_v.weight q4_K +blk.17.attn_q.weight iq4_nl +blk.17.attn_v.weight q5_0 +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_gate_exps.weight iq4_nl +blk.17.ffn_up_exps.weight iq4_nl +blk.18.attn_k.weight iq4_nl blk.18.attn_output.weight iq2_xxs -blk.18.attn_v.weight q4_K +blk.18.attn_q.weight iq4_nl +blk.18.attn_v.weight q5_0 +blk.18.ffn_down_exps.weight iq4_nl +blk.18.ffn_gate_exps.weight iq4_nl +blk.18.ffn_up_exps.weight iq4_nl +blk.19.attn_k.weight iq4_nl blk.19.attn_output.weight iq2_xxs -blk.19.attn_v.weight q4_K +blk.19.attn_q.weight iq4_nl +blk.19.attn_v.weight q5_0 +blk.19.ffn_down_exps.weight iq4_nl +blk.19.ffn_gate_exps.weight iq4_nl +blk.19.ffn_up_exps.weight iq4_nl +blk.20.attn_k.weight iq4_nl blk.20.attn_output.weight iq2_xxs -blk.20.attn_v.weight q4_K +blk.20.attn_q.weight iq4_nl +blk.20.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_gate_exps.weight iq4_nl +blk.20.ffn_up_exps.weight iq4_nl +blk.21.attn_k.weight iq4_nl blk.21.attn_output.weight iq2_xxs -blk.21.attn_v.weight q4_K +blk.21.attn_q.weight iq4_nl +blk.21.attn_v.weight q5_0 +blk.21.ffn_down_exps.weight iq4_nl +blk.21.ffn_gate_exps.weight iq4_nl +blk.21.ffn_up_exps.weight iq4_nl +blk.22.attn_k.weight iq4_nl blk.22.attn_output.weight iq2_xxs -blk.22.attn_v.weight q4_K +blk.22.attn_q.weight iq4_nl +blk.22.attn_v.weight q5_0 +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_gate_exps.weight iq4_nl +blk.22.ffn_up_exps.weight iq4_nl +blk.23.attn_k.weight iq4_nl blk.23.attn_output.weight iq2_xxs -blk.23.attn_v.weight q4_K +blk.23.attn_q.weight iq4_nl +blk.23.attn_v.weight q5_0 +blk.23.ffn_down_exps.weight iq4_nl +blk.23.ffn_gate_exps.weight iq4_nl +blk.23.ffn_up_exps.weight iq4_nl +blk.24.attn_k.weight iq4_nl blk.24.attn_output.weight iq2_xxs -blk.24.attn_v.weight q4_K +blk.24.attn_q.weight iq4_nl +blk.24.attn_v.weight q5_0 +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_gate_exps.weight iq4_nl +blk.24.ffn_up_exps.weight iq4_nl +blk.25.attn_k.weight iq4_nl blk.25.attn_output.weight iq2_xxs -blk.25.attn_v.weight q4_K +blk.25.attn_q.weight iq4_nl +blk.25.attn_v.weight q5_0 +blk.25.ffn_down_exps.weight iq4_nl +blk.25.ffn_gate_exps.weight iq4_nl +blk.25.ffn_up_exps.weight iq4_nl +blk.26.attn_k.weight iq4_nl blk.26.attn_output.weight iq2_xxs -blk.26.attn_v.weight q4_K +blk.26.attn_q.weight iq4_nl +blk.26.attn_v.weight q5_0 +blk.26.ffn_down_exps.weight iq4_nl +blk.26.ffn_gate_exps.weight iq4_nl +blk.26.ffn_up_exps.weight iq4_nl +blk.27.attn_k.weight iq4_nl blk.27.attn_output.weight iq2_xxs -blk.27.attn_v.weight q4_K +blk.27.attn_q.weight iq4_nl +blk.27.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_gate_exps.weight iq4_nl +blk.27.ffn_up_exps.weight iq4_nl +blk.28.attn_k.weight iq4_nl blk.28.attn_output.weight iq2_xxs -blk.28.attn_v.weight q4_K +blk.28.attn_q.weight iq4_nl +blk.28.attn_v.weight q5_0 +blk.28.ffn_down_exps.weight iq4_nl +blk.28.ffn_gate_exps.weight iq4_nl +blk.28.ffn_up_exps.weight iq4_nl +blk.29.attn_k.weight iq4_nl blk.29.attn_output.weight iq2_xxs -blk.29.attn_v.weight q4_K +blk.29.attn_q.weight iq4_nl +blk.29.attn_v.weight q5_0 +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_gate_exps.weight iq4_nl +blk.29.ffn_up_exps.weight iq4_nl +blk.30.attn_k.weight iq4_nl blk.30.attn_output.weight iq2_xxs -blk.30.attn_v.weight q4_K +blk.30.attn_q.weight iq4_nl +blk.30.attn_v.weight q5_0 +blk.30.ffn_down_exps.weight iq4_nl +blk.30.ffn_gate_exps.weight iq4_nl +blk.30.ffn_up_exps.weight iq4_nl +blk.31.attn_k.weight iq4_nl blk.31.attn_output.weight iq2_xxs -blk.31.attn_v.weight q4_K +blk.31.attn_q.weight iq4_nl +blk.31.attn_v.weight q5_0 +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_gate_exps.weight iq4_nl +blk.31.ffn_up_exps.weight iq4_nl +blk.32.attn_k.weight iq4_nl blk.32.attn_output.weight iq2_xxs -blk.32.attn_v.weight q4_K +blk.32.attn_q.weight iq4_nl +blk.32.attn_v.weight q5_0 +blk.32.ffn_down_exps.weight iq4_nl +blk.32.ffn_gate_exps.weight iq4_nl +blk.32.ffn_up_exps.weight iq4_nl +blk.33.attn_k.weight iq4_nl blk.33.attn_output.weight iq2_xxs -blk.33.attn_v.weight q4_K +blk.33.attn_q.weight iq4_nl +blk.33.attn_v.weight q5_0 +blk.33.ffn_down_exps.weight iq4_nl +blk.33.ffn_gate_exps.weight iq4_nl +blk.33.ffn_up_exps.weight iq4_nl +blk.34.attn_k.weight iq4_nl blk.34.attn_output.weight iq2_xxs -blk.34.attn_v.weight q4_K +blk.34.attn_q.weight iq4_nl +blk.34.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_gate_exps.weight iq4_nl +blk.34.ffn_up_exps.weight iq4_nl +blk.35.attn_k.weight iq4_nl blk.35.attn_output.weight iq2_xxs -blk.35.attn_v.weight q4_K +blk.35.attn_q.weight iq4_nl +blk.35.attn_v.weight q5_0 +blk.35.ffn_down_exps.weight iq4_nl +blk.35.ffn_gate_exps.weight iq4_nl +blk.35.ffn_up_exps.weight iq4_nl [BF16] bf16 -output.weight q6_K [TQ1_0] tq1_0 output.weight q8_0 -token_embd.weight q4_K +token_embd.weight q5_0 +blk.0.attn_k.weight q4_0 +blk.0.attn_q.weight q4_0 +blk.0.attn_v.weight q4_0 +blk.0.ffn_down_exps.weight q4_0 +blk.0.ffn_gate_exps.weight q4_0 +blk.0.ffn_up_exps.weight q4_0 +blk.1.attn_k.weight q4_0 +blk.1.attn_q.weight q4_0 +blk.1.attn_v.weight q4_0 +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_gate_exps.weight q4_0 +blk.1.ffn_up_exps.weight q4_0 +blk.2.attn_k.weight q4_0 +blk.2.attn_q.weight q4_0 +blk.2.attn_v.weight q4_0 +blk.2.ffn_down_exps.weight q4_0 +blk.2.ffn_gate_exps.weight q4_0 +blk.2.ffn_up_exps.weight q4_0 +blk.3.attn_k.weight q4_0 +blk.3.attn_q.weight q4_0 +blk.3.attn_v.weight q4_0 +blk.3.ffn_down_exps.weight q4_0 +blk.3.ffn_gate_exps.weight q4_0 +blk.3.ffn_up_exps.weight q4_0 +blk.4.attn_k.weight q4_0 +blk.4.attn_q.weight q4_0 +blk.4.attn_v.weight q4_0 +blk.4.ffn_down_exps.weight q4_0 +blk.4.ffn_gate_exps.weight q4_0 +blk.4.ffn_up_exps.weight q4_0 +blk.5.attn_k.weight q4_0 +blk.5.attn_q.weight q4_0 +blk.5.attn_v.weight q4_0 +blk.5.ffn_down_exps.weight q4_0 +blk.5.ffn_gate_exps.weight q4_0 +blk.5.ffn_up_exps.weight q4_0 +blk.6.attn_k.weight q4_0 +blk.6.attn_q.weight q4_0 +blk.6.attn_v.weight q4_0 +blk.6.ffn_down_exps.weight q4_0 +blk.6.ffn_gate_exps.weight q4_0 +blk.6.ffn_up_exps.weight q4_0 +blk.7.attn_k.weight q4_0 +blk.7.attn_q.weight q4_0 +blk.7.attn_v.weight q4_0 +blk.7.ffn_down_exps.weight q4_0 +blk.7.ffn_gate_exps.weight q4_0 +blk.7.ffn_up_exps.weight q4_0 +blk.8.attn_k.weight q4_0 +blk.8.attn_q.weight q4_0 +blk.8.attn_v.weight q4_0 +blk.8.ffn_down_exps.weight q4_0 +blk.8.ffn_gate_exps.weight q4_0 +blk.8.ffn_up_exps.weight q4_0 +blk.9.attn_k.weight q4_0 +blk.9.attn_q.weight q4_0 +blk.9.attn_v.weight q4_0 +blk.9.ffn_down_exps.weight q4_0 +blk.9.ffn_gate_exps.weight q4_0 +blk.9.ffn_up_exps.weight q4_0 +blk.10.attn_k.weight q4_0 +blk.10.attn_q.weight q4_0 +blk.10.attn_v.weight q4_0 +blk.10.ffn_down_exps.weight q4_0 +blk.10.ffn_gate_exps.weight q4_0 +blk.10.ffn_up_exps.weight q4_0 +blk.11.attn_k.weight q4_0 +blk.11.attn_q.weight q4_0 +blk.11.attn_v.weight q4_0 +blk.11.ffn_down_exps.weight q4_0 +blk.11.ffn_gate_exps.weight q4_0 +blk.11.ffn_up_exps.weight q4_0 +blk.12.attn_k.weight q4_0 +blk.12.attn_q.weight q4_0 +blk.12.attn_v.weight q4_0 +blk.12.ffn_down_exps.weight q4_0 +blk.12.ffn_gate_exps.weight q4_0 +blk.12.ffn_up_exps.weight q4_0 +blk.13.attn_k.weight q4_0 +blk.13.attn_q.weight q4_0 +blk.13.attn_v.weight q4_0 +blk.13.ffn_down_exps.weight q4_0 +blk.13.ffn_gate_exps.weight q4_0 +blk.13.ffn_up_exps.weight q4_0 +blk.14.attn_k.weight q4_0 +blk.14.attn_q.weight q4_0 +blk.14.attn_v.weight q4_0 +blk.14.ffn_down_exps.weight q4_0 +blk.14.ffn_gate_exps.weight q4_0 +blk.14.ffn_up_exps.weight q4_0 +blk.15.attn_k.weight q4_0 +blk.15.attn_q.weight q4_0 +blk.15.attn_v.weight q4_0 +blk.15.ffn_down_exps.weight q4_0 +blk.15.ffn_gate_exps.weight q4_0 +blk.15.ffn_up_exps.weight q4_0 +blk.16.attn_k.weight q4_0 +blk.16.attn_q.weight q4_0 +blk.16.attn_v.weight q4_0 +blk.16.ffn_down_exps.weight q4_0 +blk.16.ffn_gate_exps.weight q4_0 +blk.16.ffn_up_exps.weight q4_0 +blk.17.attn_k.weight q4_0 +blk.17.attn_q.weight q4_0 +blk.17.attn_v.weight q4_0 +blk.17.ffn_down_exps.weight q4_0 +blk.17.ffn_gate_exps.weight q4_0 +blk.17.ffn_up_exps.weight q4_0 +blk.18.attn_k.weight q4_0 +blk.18.attn_q.weight q4_0 +blk.18.attn_v.weight q4_0 +blk.18.ffn_down_exps.weight q4_0 +blk.18.ffn_gate_exps.weight q4_0 +blk.18.ffn_up_exps.weight q4_0 +blk.19.attn_k.weight q4_0 +blk.19.attn_q.weight q4_0 +blk.19.attn_v.weight q4_0 +blk.19.ffn_down_exps.weight q4_0 +blk.19.ffn_gate_exps.weight q4_0 +blk.19.ffn_up_exps.weight q4_0 +blk.20.attn_k.weight q4_0 +blk.20.attn_q.weight q4_0 +blk.20.attn_v.weight q4_0 +blk.20.ffn_down_exps.weight q4_0 +blk.20.ffn_gate_exps.weight q4_0 +blk.20.ffn_up_exps.weight q4_0 +blk.21.attn_k.weight q4_0 +blk.21.attn_q.weight q4_0 +blk.21.attn_v.weight q4_0 +blk.21.ffn_down_exps.weight q4_0 +blk.21.ffn_gate_exps.weight q4_0 +blk.21.ffn_up_exps.weight q4_0 +blk.22.attn_k.weight q4_0 +blk.22.attn_q.weight q4_0 +blk.22.attn_v.weight q4_0 +blk.22.ffn_down_exps.weight q4_0 +blk.22.ffn_gate_exps.weight q4_0 +blk.22.ffn_up_exps.weight q4_0 +blk.23.attn_k.weight q4_0 +blk.23.attn_q.weight q4_0 +blk.23.attn_v.weight q4_0 +blk.23.ffn_down_exps.weight q4_0 +blk.23.ffn_gate_exps.weight q4_0 +blk.23.ffn_up_exps.weight q4_0 +blk.24.attn_k.weight q4_0 +blk.24.attn_q.weight q4_0 +blk.24.attn_v.weight q4_0 +blk.24.ffn_down_exps.weight q4_0 +blk.24.ffn_gate_exps.weight q4_0 +blk.24.ffn_up_exps.weight q4_0 +blk.25.attn_k.weight q4_0 +blk.25.attn_q.weight q4_0 +blk.25.attn_v.weight q4_0 +blk.25.ffn_down_exps.weight q4_0 +blk.25.ffn_gate_exps.weight q4_0 +blk.25.ffn_up_exps.weight q4_0 +blk.26.attn_k.weight q4_0 +blk.26.attn_q.weight q4_0 +blk.26.attn_v.weight q4_0 +blk.26.ffn_down_exps.weight q4_0 +blk.26.ffn_gate_exps.weight q4_0 +blk.26.ffn_up_exps.weight q4_0 +blk.27.attn_k.weight q4_0 +blk.27.attn_q.weight q4_0 +blk.27.attn_v.weight q4_0 +blk.27.ffn_down_exps.weight q4_0 +blk.27.ffn_gate_exps.weight q4_0 +blk.27.ffn_up_exps.weight q4_0 +blk.28.attn_k.weight q4_0 +blk.28.attn_q.weight q4_0 +blk.28.attn_v.weight q4_0 +blk.28.ffn_down_exps.weight q4_0 +blk.28.ffn_gate_exps.weight q4_0 +blk.28.ffn_up_exps.weight q4_0 +blk.29.attn_k.weight q4_0 +blk.29.attn_q.weight q4_0 +blk.29.attn_v.weight q4_0 +blk.29.ffn_down_exps.weight q4_0 +blk.29.ffn_gate_exps.weight q4_0 +blk.29.ffn_up_exps.weight q4_0 +blk.30.attn_k.weight q4_0 +blk.30.attn_q.weight q4_0 +blk.30.attn_v.weight q4_0 +blk.30.ffn_down_exps.weight q4_0 +blk.30.ffn_gate_exps.weight q4_0 +blk.30.ffn_up_exps.weight q4_0 +blk.31.attn_k.weight q4_0 +blk.31.attn_q.weight q4_0 +blk.31.attn_v.weight q4_0 +blk.31.ffn_down_exps.weight q4_0 +blk.31.ffn_gate_exps.weight q4_0 +blk.31.ffn_up_exps.weight q4_0 +blk.32.attn_k.weight q4_0 +blk.32.attn_q.weight q4_0 +blk.32.attn_v.weight q4_0 +blk.32.ffn_down_exps.weight q4_0 +blk.32.ffn_gate_exps.weight q4_0 +blk.32.ffn_up_exps.weight q4_0 +blk.33.attn_k.weight q4_0 +blk.33.attn_q.weight q4_0 +blk.33.attn_v.weight q4_0 +blk.33.ffn_down_exps.weight q4_0 +blk.33.ffn_gate_exps.weight q4_0 +blk.33.ffn_up_exps.weight q4_0 +blk.34.attn_k.weight q4_0 +blk.34.attn_q.weight q4_0 +blk.34.attn_v.weight q4_0 +blk.34.ffn_down_exps.weight q4_0 +blk.34.ffn_gate_exps.weight q4_0 +blk.34.ffn_up_exps.weight q4_0 +blk.35.attn_k.weight q4_0 +blk.35.attn_q.weight q4_0 +blk.35.attn_v.weight q4_0 +blk.35.ffn_down_exps.weight q4_0 +blk.35.ffn_gate_exps.weight q4_0 +blk.35.ffn_up_exps.weight q4_0 [TQ2_0] tq2_0 output.weight q8_0 -token_embd.weight q4_K +token_embd.weight q5_0 +blk.0.attn_k.weight q4_0 +blk.0.attn_q.weight q4_0 +blk.0.attn_v.weight q4_0 +blk.0.ffn_down_exps.weight q4_0 +blk.0.ffn_gate_exps.weight q4_0 +blk.0.ffn_up_exps.weight q4_0 +blk.1.attn_k.weight q4_0 +blk.1.attn_q.weight q4_0 +blk.1.attn_v.weight q4_0 +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_gate_exps.weight q4_0 +blk.1.ffn_up_exps.weight q4_0 +blk.2.attn_k.weight q4_0 +blk.2.attn_q.weight q4_0 +blk.2.attn_v.weight q4_0 +blk.2.ffn_down_exps.weight q4_0 +blk.2.ffn_gate_exps.weight q4_0 +blk.2.ffn_up_exps.weight q4_0 +blk.3.attn_k.weight q4_0 +blk.3.attn_q.weight q4_0 +blk.3.attn_v.weight q4_0 +blk.3.ffn_down_exps.weight q4_0 +blk.3.ffn_gate_exps.weight q4_0 +blk.3.ffn_up_exps.weight q4_0 +blk.4.attn_k.weight q4_0 +blk.4.attn_q.weight q4_0 +blk.4.attn_v.weight q4_0 +blk.4.ffn_down_exps.weight q4_0 +blk.4.ffn_gate_exps.weight q4_0 +blk.4.ffn_up_exps.weight q4_0 +blk.5.attn_k.weight q4_0 +blk.5.attn_q.weight q4_0 +blk.5.attn_v.weight q4_0 +blk.5.ffn_down_exps.weight q4_0 +blk.5.ffn_gate_exps.weight q4_0 +blk.5.ffn_up_exps.weight q4_0 +blk.6.attn_k.weight q4_0 +blk.6.attn_q.weight q4_0 +blk.6.attn_v.weight q4_0 +blk.6.ffn_down_exps.weight q4_0 +blk.6.ffn_gate_exps.weight q4_0 +blk.6.ffn_up_exps.weight q4_0 +blk.7.attn_k.weight q4_0 +blk.7.attn_q.weight q4_0 +blk.7.attn_v.weight q4_0 +blk.7.ffn_down_exps.weight q4_0 +blk.7.ffn_gate_exps.weight q4_0 +blk.7.ffn_up_exps.weight q4_0 +blk.8.attn_k.weight q4_0 +blk.8.attn_q.weight q4_0 +blk.8.attn_v.weight q4_0 +blk.8.ffn_down_exps.weight q4_0 +blk.8.ffn_gate_exps.weight q4_0 +blk.8.ffn_up_exps.weight q4_0 +blk.9.attn_k.weight q4_0 +blk.9.attn_q.weight q4_0 +blk.9.attn_v.weight q4_0 +blk.9.ffn_down_exps.weight q4_0 +blk.9.ffn_gate_exps.weight q4_0 +blk.9.ffn_up_exps.weight q4_0 +blk.10.attn_k.weight q4_0 +blk.10.attn_q.weight q4_0 +blk.10.attn_v.weight q4_0 +blk.10.ffn_down_exps.weight q4_0 +blk.10.ffn_gate_exps.weight q4_0 +blk.10.ffn_up_exps.weight q4_0 +blk.11.attn_k.weight q4_0 +blk.11.attn_q.weight q4_0 +blk.11.attn_v.weight q4_0 +blk.11.ffn_down_exps.weight q4_0 +blk.11.ffn_gate_exps.weight q4_0 +blk.11.ffn_up_exps.weight q4_0 +blk.12.attn_k.weight q4_0 +blk.12.attn_q.weight q4_0 +blk.12.attn_v.weight q4_0 +blk.12.ffn_down_exps.weight q4_0 +blk.12.ffn_gate_exps.weight q4_0 +blk.12.ffn_up_exps.weight q4_0 +blk.13.attn_k.weight q4_0 +blk.13.attn_q.weight q4_0 +blk.13.attn_v.weight q4_0 +blk.13.ffn_down_exps.weight q4_0 +blk.13.ffn_gate_exps.weight q4_0 +blk.13.ffn_up_exps.weight q4_0 +blk.14.attn_k.weight q4_0 +blk.14.attn_q.weight q4_0 +blk.14.attn_v.weight q4_0 +blk.14.ffn_down_exps.weight q4_0 +blk.14.ffn_gate_exps.weight q4_0 +blk.14.ffn_up_exps.weight q4_0 +blk.15.attn_k.weight q4_0 +blk.15.attn_q.weight q4_0 +blk.15.attn_v.weight q4_0 +blk.15.ffn_down_exps.weight q4_0 +blk.15.ffn_gate_exps.weight q4_0 +blk.15.ffn_up_exps.weight q4_0 +blk.16.attn_k.weight q4_0 +blk.16.attn_q.weight q4_0 +blk.16.attn_v.weight q4_0 +blk.16.ffn_down_exps.weight q4_0 +blk.16.ffn_gate_exps.weight q4_0 +blk.16.ffn_up_exps.weight q4_0 +blk.17.attn_k.weight q4_0 +blk.17.attn_q.weight q4_0 +blk.17.attn_v.weight q4_0 +blk.17.ffn_down_exps.weight q4_0 +blk.17.ffn_gate_exps.weight q4_0 +blk.17.ffn_up_exps.weight q4_0 +blk.18.attn_k.weight q4_0 +blk.18.attn_q.weight q4_0 +blk.18.attn_v.weight q4_0 +blk.18.ffn_down_exps.weight q4_0 +blk.18.ffn_gate_exps.weight q4_0 +blk.18.ffn_up_exps.weight q4_0 +blk.19.attn_k.weight q4_0 +blk.19.attn_q.weight q4_0 +blk.19.attn_v.weight q4_0 +blk.19.ffn_down_exps.weight q4_0 +blk.19.ffn_gate_exps.weight q4_0 +blk.19.ffn_up_exps.weight q4_0 +blk.20.attn_k.weight q4_0 +blk.20.attn_q.weight q4_0 +blk.20.attn_v.weight q4_0 +blk.20.ffn_down_exps.weight q4_0 +blk.20.ffn_gate_exps.weight q4_0 +blk.20.ffn_up_exps.weight q4_0 +blk.21.attn_k.weight q4_0 +blk.21.attn_q.weight q4_0 +blk.21.attn_v.weight q4_0 +blk.21.ffn_down_exps.weight q4_0 +blk.21.ffn_gate_exps.weight q4_0 +blk.21.ffn_up_exps.weight q4_0 +blk.22.attn_k.weight q4_0 +blk.22.attn_q.weight q4_0 +blk.22.attn_v.weight q4_0 +blk.22.ffn_down_exps.weight q4_0 +blk.22.ffn_gate_exps.weight q4_0 +blk.22.ffn_up_exps.weight q4_0 +blk.23.attn_k.weight q4_0 +blk.23.attn_q.weight q4_0 +blk.23.attn_v.weight q4_0 +blk.23.ffn_down_exps.weight q4_0 +blk.23.ffn_gate_exps.weight q4_0 +blk.23.ffn_up_exps.weight q4_0 +blk.24.attn_k.weight q4_0 +blk.24.attn_q.weight q4_0 +blk.24.attn_v.weight q4_0 +blk.24.ffn_down_exps.weight q4_0 +blk.24.ffn_gate_exps.weight q4_0 +blk.24.ffn_up_exps.weight q4_0 +blk.25.attn_k.weight q4_0 +blk.25.attn_q.weight q4_0 +blk.25.attn_v.weight q4_0 +blk.25.ffn_down_exps.weight q4_0 +blk.25.ffn_gate_exps.weight q4_0 +blk.25.ffn_up_exps.weight q4_0 +blk.26.attn_k.weight q4_0 +blk.26.attn_q.weight q4_0 +blk.26.attn_v.weight q4_0 +blk.26.ffn_down_exps.weight q4_0 +blk.26.ffn_gate_exps.weight q4_0 +blk.26.ffn_up_exps.weight q4_0 +blk.27.attn_k.weight q4_0 +blk.27.attn_q.weight q4_0 +blk.27.attn_v.weight q4_0 +blk.27.ffn_down_exps.weight q4_0 +blk.27.ffn_gate_exps.weight q4_0 +blk.27.ffn_up_exps.weight q4_0 +blk.28.attn_k.weight q4_0 +blk.28.attn_q.weight q4_0 +blk.28.attn_v.weight q4_0 +blk.28.ffn_down_exps.weight q4_0 +blk.28.ffn_gate_exps.weight q4_0 +blk.28.ffn_up_exps.weight q4_0 +blk.29.attn_k.weight q4_0 +blk.29.attn_q.weight q4_0 +blk.29.attn_v.weight q4_0 +blk.29.ffn_down_exps.weight q4_0 +blk.29.ffn_gate_exps.weight q4_0 +blk.29.ffn_up_exps.weight q4_0 +blk.30.attn_k.weight q4_0 +blk.30.attn_q.weight q4_0 +blk.30.attn_v.weight q4_0 +blk.30.ffn_down_exps.weight q4_0 +blk.30.ffn_gate_exps.weight q4_0 +blk.30.ffn_up_exps.weight q4_0 +blk.31.attn_k.weight q4_0 +blk.31.attn_q.weight q4_0 +blk.31.attn_v.weight q4_0 +blk.31.ffn_down_exps.weight q4_0 +blk.31.ffn_gate_exps.weight q4_0 +blk.31.ffn_up_exps.weight q4_0 +blk.32.attn_k.weight q4_0 +blk.32.attn_q.weight q4_0 +blk.32.attn_v.weight q4_0 +blk.32.ffn_down_exps.weight q4_0 +blk.32.ffn_gate_exps.weight q4_0 +blk.32.ffn_up_exps.weight q4_0 +blk.33.attn_k.weight q4_0 +blk.33.attn_q.weight q4_0 +blk.33.attn_v.weight q4_0 +blk.33.ffn_down_exps.weight q4_0 +blk.33.ffn_gate_exps.weight q4_0 +blk.33.ffn_up_exps.weight q4_0 +blk.34.attn_k.weight q4_0 +blk.34.attn_q.weight q4_0 +blk.34.attn_v.weight q4_0 +blk.34.ffn_down_exps.weight q4_0 +blk.34.ffn_gate_exps.weight q4_0 +blk.34.ffn_up_exps.weight q4_0 +blk.35.attn_k.weight q4_0 +blk.35.attn_q.weight q4_0 +blk.35.attn_v.weight q4_0 +blk.35.ffn_down_exps.weight q4_0 +blk.35.ffn_gate_exps.weight q4_0 +blk.35.ffn_up_exps.weight q4_0 [MXFP4_MOE] mxfp4 output.weight q8_0 diff --git a/tests/snapshots/meta-llama-3.1-70b-instruct.schema b/tests/snapshots/meta-llama-3.1-70b-instruct.schema index b26755d6f7..95e0fdf148 100644 --- a/tests/snapshots/meta-llama-3.1-70b-instruct.schema +++ b/tests/snapshots/meta-llama-3.1-70b-instruct.schema @@ -2,10 +2,8 @@ # n_embd=8192, n_ff=28672, n_vocab=128256, n_layer=80, n_head=64, n_head_kv=8 [F32] f32 -output.weight q6_K [F16] f16 -output.weight q6_K [Q4_0] q4_0 output.weight q6_K @@ -3324,7 +3322,6 @@ blk.79.attn_v.weight q4_K output.weight q5_K [BF16] bf16 -output.weight q6_K [TQ1_0] tq1_0 token_embd.weight q4_K diff --git a/tests/snapshots/nemotron-nano-3-30b-a3b.schema b/tests/snapshots/nemotron-nano-3-30b-a3b.schema index c68a6e0892..cf46f9de4b 100644 --- a/tests/snapshots/nemotron-nano-3-30b-a3b.schema +++ b/tests/snapshots/nemotron-nano-3-30b-a3b.schema @@ -2,578 +2,3236 @@ # n_embd=2688, n_ff=0, n_vocab=131072, n_layer=52, n_head=32, n_head_kv=0, n_expert=128 [F32] f32 -output.weight q6_K [F16] f16 -output.weight q6_K [Q4_0] q4_0 -output.weight q6_K +output.weight q8_0 [Q4_1] q4_1 -output.weight q6_K +output.weight q8_0 [Q8_0] q8_0 [Q5_0] q5_0 -output.weight q6_K +output.weight q8_0 [Q5_1] q5_1 -output.weight q6_K +output.weight q8_0 [Q2_K] q2_K output.weight q8_0 -blk.1.ffn_down_exps.weight q3_K -blk.1.ffn_down_shexp.weight q3_K -blk.3.ffn_down_exps.weight q3_K -blk.3.ffn_down_shexp.weight q3_K +token_embd.weight q4_0 +blk.0.ssm_in.weight q4_0 +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_down_shexp.weight q4_0 +blk.1.ffn_up_exps.weight q4_0 +blk.1.ffn_up_shexp.weight q4_0 +blk.2.ssm_in.weight q4_0 +blk.3.ffn_down_exps.weight q4_0 +blk.3.ffn_down_shexp.weight q4_0 +blk.3.ffn_up_exps.weight q4_0 +blk.3.ffn_up_shexp.weight q4_0 +blk.4.ssm_in.weight q4_0 +blk.5.attn_k.weight q4_0 blk.5.attn_output.weight q3_K -blk.5.attn_v.weight q3_K -blk.6.ffn_down_exps.weight q3_K -blk.6.ffn_down_shexp.weight q3_K -blk.8.ffn_down_exps.weight q3_K -blk.8.ffn_down_shexp.weight q3_K -blk.10.ffn_down_exps.weight q3_K -blk.10.ffn_down_shexp.weight q3_K +blk.5.attn_q.weight q4_0 +blk.5.attn_v.weight q4_0 +blk.6.ffn_down_exps.weight q4_0 +blk.6.ffn_down_shexp.weight q4_0 +blk.6.ffn_up_exps.weight q4_0 +blk.6.ffn_up_shexp.weight q4_0 +blk.7.ssm_in.weight q4_0 +blk.8.ffn_down_exps.weight q4_0 +blk.8.ffn_down_shexp.weight q4_0 +blk.8.ffn_up_exps.weight q4_0 +blk.8.ffn_up_shexp.weight q4_0 +blk.9.ssm_in.weight q4_0 +blk.10.ffn_down_exps.weight q4_0 +blk.10.ffn_down_shexp.weight q4_0 +blk.10.ffn_up_exps.weight q4_0 +blk.10.ffn_up_shexp.weight q4_0 +blk.11.ssm_in.weight q4_0 +blk.12.attn_k.weight q4_0 blk.12.attn_output.weight q3_K -blk.12.attn_v.weight q3_K -blk.13.ffn_down_exps.weight q3_K -blk.13.ffn_down_shexp.weight q3_K -blk.15.ffn_down_exps.weight q3_K -blk.15.ffn_down_shexp.weight q3_K -blk.17.ffn_down_exps.weight q3_K -blk.17.ffn_down_shexp.weight q3_K +blk.12.attn_q.weight q4_0 +blk.12.attn_v.weight q4_0 +blk.13.ffn_down_exps.weight q4_0 +blk.13.ffn_down_shexp.weight q4_0 +blk.13.ffn_up_exps.weight q4_0 +blk.13.ffn_up_shexp.weight q4_0 +blk.14.ssm_in.weight q4_0 +blk.15.ffn_down_exps.weight q4_0 +blk.15.ffn_down_shexp.weight q4_0 +blk.15.ffn_up_exps.weight q4_0 +blk.15.ffn_up_shexp.weight q4_0 +blk.16.ssm_in.weight q4_0 +blk.17.ffn_down_exps.weight q4_0 +blk.17.ffn_down_shexp.weight q4_0 +blk.17.ffn_up_exps.weight q4_0 +blk.17.ffn_up_shexp.weight q4_0 +blk.18.ssm_in.weight q4_0 +blk.19.attn_k.weight q4_0 blk.19.attn_output.weight q3_K -blk.19.attn_v.weight q3_K -blk.20.ffn_down_exps.weight q3_K -blk.20.ffn_down_shexp.weight q3_K -blk.22.ffn_down_exps.weight q3_K -blk.22.ffn_down_shexp.weight q3_K -blk.24.ffn_down_exps.weight q3_K -blk.24.ffn_down_shexp.weight q3_K +blk.19.attn_q.weight q4_0 +blk.19.attn_v.weight q4_0 +blk.20.ffn_down_exps.weight q4_0 +blk.20.ffn_down_shexp.weight q4_0 +blk.20.ffn_up_exps.weight q4_0 +blk.20.ffn_up_shexp.weight q4_0 +blk.21.ssm_in.weight q4_0 +blk.22.ffn_down_exps.weight q4_0 +blk.22.ffn_down_shexp.weight q4_0 +blk.22.ffn_up_exps.weight q4_0 +blk.22.ffn_up_shexp.weight q4_0 +blk.23.ssm_in.weight q4_0 +blk.24.ffn_down_exps.weight q4_0 +blk.24.ffn_down_shexp.weight q4_0 +blk.24.ffn_up_exps.weight q4_0 +blk.24.ffn_up_shexp.weight q4_0 +blk.25.ssm_in.weight q4_0 +blk.26.attn_k.weight q4_0 blk.26.attn_output.weight q3_K -blk.26.attn_v.weight q3_K -blk.27.ffn_down_exps.weight q3_K -blk.27.ffn_down_shexp.weight q3_K -blk.29.ffn_down_exps.weight q3_K -blk.29.ffn_down_shexp.weight q3_K -blk.31.ffn_down_exps.weight q3_K -blk.31.ffn_down_shexp.weight q3_K +blk.26.attn_q.weight q4_0 +blk.26.attn_v.weight q4_0 +blk.27.ffn_down_exps.weight q4_0 +blk.27.ffn_down_shexp.weight q4_0 +blk.27.ffn_up_exps.weight q4_0 +blk.27.ffn_up_shexp.weight q4_0 +blk.28.ssm_in.weight q4_0 +blk.29.ffn_down_exps.weight q4_0 +blk.29.ffn_down_shexp.weight q4_0 +blk.29.ffn_up_exps.weight q4_0 +blk.29.ffn_up_shexp.weight q4_0 +blk.30.ssm_in.weight q4_0 +blk.31.ffn_down_exps.weight q4_0 +blk.31.ffn_down_shexp.weight q4_0 +blk.31.ffn_up_exps.weight q4_0 +blk.31.ffn_up_shexp.weight q4_0 +blk.32.ssm_in.weight q4_0 +blk.33.attn_k.weight q4_0 blk.33.attn_output.weight q3_K -blk.33.attn_v.weight q3_K -blk.34.ffn_down_exps.weight q3_K -blk.34.ffn_down_shexp.weight q3_K -blk.36.ffn_down_exps.weight q3_K -blk.36.ffn_down_shexp.weight q3_K -blk.38.ffn_down_exps.weight q3_K -blk.38.ffn_down_shexp.weight q3_K -blk.40.ffn_down_exps.weight q3_K -blk.40.ffn_down_shexp.weight q3_K +blk.33.attn_q.weight q4_0 +blk.33.attn_v.weight q4_0 +blk.34.ffn_down_exps.weight q4_0 +blk.34.ffn_down_shexp.weight q4_0 +blk.34.ffn_up_exps.weight q4_0 +blk.34.ffn_up_shexp.weight q4_0 +blk.35.ssm_in.weight q4_0 +blk.36.ffn_down_exps.weight q4_0 +blk.36.ffn_down_shexp.weight q4_0 +blk.36.ffn_up_exps.weight q4_0 +blk.36.ffn_up_shexp.weight q4_0 +blk.37.ssm_in.weight q4_0 +blk.38.ffn_down_exps.weight q4_0 +blk.38.ffn_down_shexp.weight q4_0 +blk.38.ffn_up_exps.weight q4_0 +blk.38.ffn_up_shexp.weight q4_0 +blk.39.ssm_in.weight q4_0 +blk.40.ffn_down_exps.weight q4_0 +blk.40.ffn_down_shexp.weight q4_0 +blk.40.ffn_up_exps.weight q4_0 +blk.40.ffn_up_shexp.weight q4_0 +blk.41.ssm_in.weight q4_0 +blk.42.attn_k.weight q4_0 blk.42.attn_output.weight q3_K -blk.42.attn_v.weight q3_K -blk.43.ffn_down_exps.weight q3_K -blk.43.ffn_down_shexp.weight q3_K -blk.45.ffn_down_exps.weight q3_K -blk.45.ffn_down_shexp.weight q3_K -blk.47.ffn_down_exps.weight q3_K -blk.47.ffn_down_shexp.weight q3_K -blk.49.ffn_down_exps.weight q3_K -blk.49.ffn_down_shexp.weight q3_K -blk.51.ffn_down_exps.weight q3_K -blk.51.ffn_down_shexp.weight q3_K +blk.42.attn_q.weight q4_0 +blk.42.attn_v.weight q4_0 +blk.43.ffn_down_exps.weight q4_0 +blk.43.ffn_down_shexp.weight q4_0 +blk.43.ffn_up_exps.weight q4_0 +blk.43.ffn_up_shexp.weight q4_0 +blk.44.ssm_in.weight q4_0 +blk.45.ffn_down_exps.weight q4_0 +blk.45.ffn_down_shexp.weight q4_0 +blk.45.ffn_up_exps.weight q4_0 +blk.45.ffn_up_shexp.weight q4_0 +blk.46.ssm_in.weight q4_0 +blk.47.ffn_down_exps.weight q4_0 +blk.47.ffn_down_shexp.weight q4_0 +blk.47.ffn_up_exps.weight q4_0 +blk.47.ffn_up_shexp.weight q4_0 +blk.48.ssm_in.weight q4_0 +blk.49.ffn_down_exps.weight q4_0 +blk.49.ffn_down_shexp.weight q4_0 +blk.49.ffn_up_exps.weight q4_0 +blk.49.ffn_up_shexp.weight q4_0 +blk.50.ssm_in.weight q4_0 +blk.51.ffn_down_exps.weight q4_0 +blk.51.ffn_down_shexp.weight q4_0 +blk.51.ffn_up_exps.weight q4_0 +blk.51.ffn_up_shexp.weight q4_0 [Q3_K_S] q3_K output.weight q8_0 +token_embd.weight q4_0 +blk.0.ssm_in.weight q4_0 +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_down_shexp.weight q4_0 +blk.1.ffn_up_exps.weight q4_0 +blk.1.ffn_up_shexp.weight q4_0 +blk.2.ssm_in.weight q4_0 +blk.3.ffn_down_exps.weight q4_0 +blk.3.ffn_down_shexp.weight q4_0 +blk.3.ffn_up_exps.weight q4_0 +blk.3.ffn_up_shexp.weight q4_0 +blk.4.ssm_in.weight q4_0 +blk.5.attn_k.weight q4_0 +blk.5.attn_q.weight q4_0 +blk.5.attn_v.weight q4_0 +blk.6.ffn_down_exps.weight q4_0 +blk.6.ffn_down_shexp.weight q4_0 +blk.6.ffn_up_exps.weight q4_0 +blk.6.ffn_up_shexp.weight q4_0 +blk.7.ssm_in.weight q4_0 +blk.8.ffn_down_exps.weight q4_0 +blk.8.ffn_down_shexp.weight q4_0 +blk.8.ffn_up_exps.weight q4_0 +blk.8.ffn_up_shexp.weight q4_0 +blk.9.ssm_in.weight q4_0 +blk.10.ffn_down_exps.weight q4_0 +blk.10.ffn_down_shexp.weight q4_0 +blk.10.ffn_up_exps.weight q4_0 +blk.10.ffn_up_shexp.weight q4_0 +blk.11.ssm_in.weight q4_0 +blk.12.attn_k.weight q4_0 +blk.12.attn_q.weight q4_0 +blk.12.attn_v.weight q4_0 +blk.13.ffn_down_exps.weight q4_0 +blk.13.ffn_down_shexp.weight q4_0 +blk.13.ffn_up_exps.weight q4_0 +blk.13.ffn_up_shexp.weight q4_0 +blk.14.ssm_in.weight q4_0 +blk.15.ffn_down_exps.weight q4_0 +blk.15.ffn_down_shexp.weight q4_0 +blk.15.ffn_up_exps.weight q4_0 +blk.15.ffn_up_shexp.weight q4_0 +blk.16.ssm_in.weight q4_0 +blk.17.ffn_down_exps.weight q4_0 +blk.17.ffn_down_shexp.weight q4_0 +blk.17.ffn_up_exps.weight q4_0 +blk.17.ffn_up_shexp.weight q4_0 +blk.18.ssm_in.weight q4_0 +blk.19.attn_k.weight q4_0 +blk.19.attn_q.weight q4_0 +blk.19.attn_v.weight q4_0 +blk.20.ffn_down_exps.weight q4_0 +blk.20.ffn_down_shexp.weight q4_0 +blk.20.ffn_up_exps.weight q4_0 +blk.20.ffn_up_shexp.weight q4_0 +blk.21.ssm_in.weight q4_0 +blk.22.ffn_down_exps.weight q4_0 +blk.22.ffn_down_shexp.weight q4_0 +blk.22.ffn_up_exps.weight q4_0 +blk.22.ffn_up_shexp.weight q4_0 +blk.23.ssm_in.weight q4_0 +blk.24.ffn_down_exps.weight q4_0 +blk.24.ffn_down_shexp.weight q4_0 +blk.24.ffn_up_exps.weight q4_0 +blk.24.ffn_up_shexp.weight q4_0 +blk.25.ssm_in.weight q4_0 +blk.26.attn_k.weight q4_0 +blk.26.attn_q.weight q4_0 +blk.26.attn_v.weight q4_0 +blk.27.ffn_down_exps.weight q4_0 +blk.27.ffn_down_shexp.weight q4_0 +blk.27.ffn_up_exps.weight q4_0 +blk.27.ffn_up_shexp.weight q4_0 +blk.28.ssm_in.weight q4_0 +blk.29.ffn_down_exps.weight q4_0 +blk.29.ffn_down_shexp.weight q4_0 +blk.29.ffn_up_exps.weight q4_0 +blk.29.ffn_up_shexp.weight q4_0 +blk.30.ssm_in.weight q4_0 +blk.31.ffn_down_exps.weight q4_0 +blk.31.ffn_down_shexp.weight q4_0 +blk.31.ffn_up_exps.weight q4_0 +blk.31.ffn_up_shexp.weight q4_0 +blk.32.ssm_in.weight q4_0 +blk.33.attn_k.weight q4_0 +blk.33.attn_q.weight q4_0 +blk.33.attn_v.weight q4_0 +blk.34.ffn_down_exps.weight q4_0 +blk.34.ffn_down_shexp.weight q4_0 +blk.34.ffn_up_exps.weight q4_0 +blk.34.ffn_up_shexp.weight q4_0 +blk.35.ssm_in.weight q4_0 +blk.36.ffn_down_exps.weight q4_0 +blk.36.ffn_down_shexp.weight q4_0 +blk.36.ffn_up_exps.weight q4_0 +blk.36.ffn_up_shexp.weight q4_0 +blk.37.ssm_in.weight q4_0 +blk.38.ffn_down_exps.weight q4_0 +blk.38.ffn_down_shexp.weight q4_0 +blk.38.ffn_up_exps.weight q4_0 +blk.38.ffn_up_shexp.weight q4_0 +blk.39.ssm_in.weight q4_0 +blk.40.ffn_down_exps.weight q4_0 +blk.40.ffn_down_shexp.weight q4_0 +blk.40.ffn_up_exps.weight q4_0 +blk.40.ffn_up_shexp.weight q4_0 +blk.41.ssm_in.weight q4_0 +blk.42.attn_k.weight q4_0 +blk.42.attn_q.weight q4_0 +blk.42.attn_v.weight q4_0 +blk.43.ffn_down_exps.weight q4_0 +blk.43.ffn_down_shexp.weight q4_0 +blk.43.ffn_up_exps.weight q4_0 +blk.43.ffn_up_shexp.weight q4_0 +blk.44.ssm_in.weight q4_0 +blk.45.ffn_down_exps.weight q4_0 +blk.45.ffn_down_shexp.weight q4_0 +blk.45.ffn_up_exps.weight q4_0 +blk.45.ffn_up_shexp.weight q4_0 +blk.46.ssm_in.weight q4_0 +blk.47.ffn_down_exps.weight q4_0 +blk.47.ffn_down_shexp.weight q4_0 +blk.47.ffn_up_exps.weight q4_0 +blk.47.ffn_up_shexp.weight q4_0 +blk.48.ssm_in.weight q4_0 +blk.49.ffn_down_exps.weight q4_0 +blk.49.ffn_down_shexp.weight q4_0 +blk.49.ffn_up_exps.weight q4_0 +blk.49.ffn_up_shexp.weight q4_0 +blk.50.ssm_in.weight q4_0 +blk.51.ffn_down_exps.weight q4_0 +blk.51.ffn_down_shexp.weight q4_0 +blk.51.ffn_up_exps.weight q4_0 +blk.51.ffn_up_shexp.weight q4_0 [Q3_K_M] q3_K output.weight q8_0 -blk.1.ffn_down_exps.weight q5_K -blk.1.ffn_down_shexp.weight q5_K -blk.3.ffn_down_exps.weight q4_K -blk.3.ffn_down_shexp.weight q4_K +token_embd.weight q4_0 +blk.0.ssm_in.weight q4_0 +blk.1.ffn_down_exps.weight q5_1 +blk.1.ffn_down_shexp.weight q5_1 +blk.1.ffn_up_exps.weight q4_0 +blk.1.ffn_up_shexp.weight q4_0 +blk.2.ssm_in.weight q4_0 +blk.3.ffn_down_exps.weight q5_0 +blk.3.ffn_down_shexp.weight q5_0 +blk.3.ffn_up_exps.weight q4_0 +blk.3.ffn_up_shexp.weight q4_0 +blk.4.ssm_in.weight q4_0 +blk.5.attn_k.weight q4_0 blk.5.attn_output.weight q4_K -blk.5.attn_v.weight q5_K -blk.6.ffn_down_exps.weight q4_K -blk.6.ffn_down_shexp.weight q4_K -blk.8.ffn_down_exps.weight q4_K -blk.8.ffn_down_shexp.weight q4_K -blk.10.ffn_down_exps.weight q4_K -blk.10.ffn_down_shexp.weight q4_K +blk.5.attn_q.weight q4_0 +blk.5.attn_v.weight q5_1 +blk.6.ffn_down_exps.weight q5_0 +blk.6.ffn_down_shexp.weight q5_0 +blk.6.ffn_up_exps.weight q4_0 +blk.6.ffn_up_shexp.weight q4_0 +blk.7.ssm_in.weight q4_0 +blk.8.ffn_down_exps.weight q5_0 +blk.8.ffn_down_shexp.weight q5_0 +blk.8.ffn_up_exps.weight q4_0 +blk.8.ffn_up_shexp.weight q4_0 +blk.9.ssm_in.weight q4_0 +blk.10.ffn_down_exps.weight q5_0 +blk.10.ffn_down_shexp.weight q5_0 +blk.10.ffn_up_exps.weight q4_0 +blk.10.ffn_up_shexp.weight q4_0 +blk.11.ssm_in.weight q4_0 +blk.12.attn_k.weight q4_0 blk.12.attn_output.weight q4_K -blk.12.attn_v.weight q5_K -blk.13.ffn_down_exps.weight q4_K -blk.13.ffn_down_shexp.weight q4_K -blk.15.ffn_down_exps.weight q4_K -blk.15.ffn_down_shexp.weight q4_K -blk.17.ffn_down_exps.weight q4_K -blk.17.ffn_down_shexp.weight q4_K +blk.12.attn_q.weight q4_0 +blk.12.attn_v.weight q5_1 +blk.13.ffn_down_exps.weight q5_0 +blk.13.ffn_down_shexp.weight q5_0 +blk.13.ffn_up_exps.weight q4_0 +blk.13.ffn_up_shexp.weight q4_0 +blk.14.ssm_in.weight q4_0 +blk.15.ffn_down_exps.weight q5_0 +blk.15.ffn_down_shexp.weight q5_0 +blk.15.ffn_up_exps.weight q4_0 +blk.15.ffn_up_shexp.weight q4_0 +blk.16.ssm_in.weight q4_0 +blk.17.ffn_down_exps.weight q5_0 +blk.17.ffn_down_shexp.weight q5_0 +blk.17.ffn_up_exps.weight q4_0 +blk.17.ffn_up_shexp.weight q4_0 +blk.18.ssm_in.weight q4_0 +blk.19.attn_k.weight q4_0 blk.19.attn_output.weight q4_K -blk.19.attn_v.weight q4_K -blk.20.ffn_down_exps.weight q4_K -blk.20.ffn_down_shexp.weight q4_K -blk.22.ffn_down_exps.weight q4_K -blk.22.ffn_down_shexp.weight q4_K -blk.24.ffn_down_exps.weight q4_K -blk.24.ffn_down_shexp.weight q4_K +blk.19.attn_q.weight q4_0 +blk.19.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight q5_0 +blk.20.ffn_down_shexp.weight q5_0 +blk.20.ffn_up_exps.weight q4_0 +blk.20.ffn_up_shexp.weight q4_0 +blk.21.ssm_in.weight q4_0 +blk.22.ffn_down_exps.weight q5_0 +blk.22.ffn_down_shexp.weight q5_0 +blk.22.ffn_up_exps.weight q4_0 +blk.22.ffn_up_shexp.weight q4_0 +blk.23.ssm_in.weight q4_0 +blk.24.ffn_down_exps.weight q5_0 +blk.24.ffn_down_shexp.weight q5_0 +blk.24.ffn_up_exps.weight q4_0 +blk.24.ffn_up_shexp.weight q4_0 +blk.25.ssm_in.weight q4_0 +blk.26.attn_k.weight q4_0 blk.26.attn_output.weight q4_K -blk.26.attn_v.weight q4_K -blk.27.ffn_down_exps.weight q4_K -blk.27.ffn_down_shexp.weight q4_K -blk.29.ffn_down_exps.weight q4_K -blk.29.ffn_down_shexp.weight q4_K -blk.31.ffn_down_exps.weight q4_K -blk.31.ffn_down_shexp.weight q4_K +blk.26.attn_q.weight q4_0 +blk.26.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight q5_0 +blk.27.ffn_down_shexp.weight q5_0 +blk.27.ffn_up_exps.weight q4_0 +blk.27.ffn_up_shexp.weight q4_0 +blk.28.ssm_in.weight q4_0 +blk.29.ffn_down_exps.weight q5_0 +blk.29.ffn_down_shexp.weight q5_0 +blk.29.ffn_up_exps.weight q4_0 +blk.29.ffn_up_shexp.weight q4_0 +blk.30.ssm_in.weight q4_0 +blk.31.ffn_down_exps.weight q5_0 +blk.31.ffn_down_shexp.weight q5_0 +blk.31.ffn_up_exps.weight q4_0 +blk.31.ffn_up_shexp.weight q4_0 +blk.32.ssm_in.weight q4_0 +blk.33.attn_k.weight q4_0 blk.33.attn_output.weight q4_K -blk.33.attn_v.weight q4_K -blk.34.ffn_down_exps.weight q4_K -blk.34.ffn_down_shexp.weight q4_K -blk.36.ffn_down_exps.weight q4_K -blk.36.ffn_down_shexp.weight q4_K -blk.38.ffn_down_exps.weight q4_K -blk.38.ffn_down_shexp.weight q4_K -blk.40.ffn_down_exps.weight q4_K -blk.40.ffn_down_shexp.weight q4_K +blk.33.attn_q.weight q4_0 +blk.33.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight q5_0 +blk.34.ffn_down_shexp.weight q5_0 +blk.34.ffn_up_exps.weight q4_0 +blk.34.ffn_up_shexp.weight q4_0 +blk.35.ssm_in.weight q4_0 +blk.36.ffn_down_exps.weight q5_0 +blk.36.ffn_down_shexp.weight q5_0 +blk.36.ffn_up_exps.weight q4_0 +blk.36.ffn_up_shexp.weight q4_0 +blk.37.ssm_in.weight q4_0 +blk.38.ffn_down_exps.weight q5_0 +blk.38.ffn_down_shexp.weight q5_0 +blk.38.ffn_up_exps.weight q4_0 +blk.38.ffn_up_shexp.weight q4_0 +blk.39.ssm_in.weight q4_0 +blk.40.ffn_down_exps.weight q5_0 +blk.40.ffn_down_shexp.weight q5_0 +blk.40.ffn_up_exps.weight q4_0 +blk.40.ffn_up_shexp.weight q4_0 +blk.41.ssm_in.weight q4_0 +blk.42.attn_k.weight q4_0 blk.42.attn_output.weight q4_K -blk.42.attn_v.weight q4_K -blk.43.ffn_down_exps.weight q4_K -blk.43.ffn_down_shexp.weight q4_K -blk.45.ffn_down_exps.weight q4_K -blk.45.ffn_down_shexp.weight q4_K -blk.47.ffn_down_exps.weight q4_K -blk.47.ffn_down_shexp.weight q4_K -blk.49.ffn_down_exps.weight q4_K -blk.49.ffn_down_shexp.weight q4_K -blk.51.ffn_down_exps.weight q4_K -blk.51.ffn_down_shexp.weight q4_K +blk.42.attn_q.weight q4_0 +blk.42.attn_v.weight q5_0 +blk.43.ffn_down_exps.weight q5_0 +blk.43.ffn_down_shexp.weight q5_0 +blk.43.ffn_up_exps.weight q4_0 +blk.43.ffn_up_shexp.weight q4_0 +blk.44.ssm_in.weight q4_0 +blk.45.ffn_down_exps.weight q5_0 +blk.45.ffn_down_shexp.weight q5_0 +blk.45.ffn_up_exps.weight q4_0 +blk.45.ffn_up_shexp.weight q4_0 +blk.46.ssm_in.weight q4_0 +blk.47.ffn_down_exps.weight q5_0 +blk.47.ffn_down_shexp.weight q5_0 +blk.47.ffn_up_exps.weight q4_0 +blk.47.ffn_up_shexp.weight q4_0 +blk.48.ssm_in.weight q4_0 +blk.49.ffn_down_exps.weight q5_0 +blk.49.ffn_down_shexp.weight q5_0 +blk.49.ffn_up_exps.weight q4_0 +blk.49.ffn_up_shexp.weight q4_0 +blk.50.ssm_in.weight q4_0 +blk.51.ffn_down_exps.weight q5_0 +blk.51.ffn_down_shexp.weight q5_0 +blk.51.ffn_up_exps.weight q4_0 +blk.51.ffn_up_shexp.weight q4_0 [Q3_K_L] q3_K output.weight q8_0 -blk.1.ffn_down_exps.weight q5_K -blk.1.ffn_down_shexp.weight q5_K -blk.3.ffn_down_exps.weight q5_K -blk.3.ffn_down_shexp.weight q5_K +token_embd.weight q4_0 +blk.0.ssm_in.weight q4_0 +blk.1.ffn_down_exps.weight q5_1 +blk.1.ffn_down_shexp.weight q5_1 +blk.1.ffn_up_exps.weight q4_0 +blk.1.ffn_up_shexp.weight q4_0 +blk.2.ssm_in.weight q4_0 +blk.3.ffn_down_exps.weight q5_1 +blk.3.ffn_down_shexp.weight q5_1 +blk.3.ffn_up_exps.weight q4_0 +blk.3.ffn_up_shexp.weight q4_0 +blk.4.ssm_in.weight q4_0 +blk.5.attn_k.weight q4_0 blk.5.attn_output.weight q5_K -blk.5.attn_v.weight q5_K -blk.6.ffn_down_exps.weight q5_K -blk.6.ffn_down_shexp.weight q5_K -blk.8.ffn_down_exps.weight q5_K -blk.8.ffn_down_shexp.weight q5_K -blk.10.ffn_down_exps.weight q5_K -blk.10.ffn_down_shexp.weight q5_K +blk.5.attn_q.weight q4_0 +blk.5.attn_v.weight q5_1 +blk.6.ffn_down_exps.weight q5_1 +blk.6.ffn_down_shexp.weight q5_1 +blk.6.ffn_up_exps.weight q4_0 +blk.6.ffn_up_shexp.weight q4_0 +blk.7.ssm_in.weight q4_0 +blk.8.ffn_down_exps.weight q5_1 +blk.8.ffn_down_shexp.weight q5_1 +blk.8.ffn_up_exps.weight q4_0 +blk.8.ffn_up_shexp.weight q4_0 +blk.9.ssm_in.weight q4_0 +blk.10.ffn_down_exps.weight q5_1 +blk.10.ffn_down_shexp.weight q5_1 +blk.10.ffn_up_exps.weight q4_0 +blk.10.ffn_up_shexp.weight q4_0 +blk.11.ssm_in.weight q4_0 +blk.12.attn_k.weight q4_0 blk.12.attn_output.weight q5_K -blk.12.attn_v.weight q5_K -blk.13.ffn_down_exps.weight q5_K -blk.13.ffn_down_shexp.weight q5_K -blk.15.ffn_down_exps.weight q5_K -blk.15.ffn_down_shexp.weight q5_K -blk.17.ffn_down_exps.weight q5_K -blk.17.ffn_down_shexp.weight q5_K +blk.12.attn_q.weight q4_0 +blk.12.attn_v.weight q5_1 +blk.13.ffn_down_exps.weight q5_1 +blk.13.ffn_down_shexp.weight q5_1 +blk.13.ffn_up_exps.weight q4_0 +blk.13.ffn_up_shexp.weight q4_0 +blk.14.ssm_in.weight q4_0 +blk.15.ffn_down_exps.weight q5_1 +blk.15.ffn_down_shexp.weight q5_1 +blk.15.ffn_up_exps.weight q4_0 +blk.15.ffn_up_shexp.weight q4_0 +blk.16.ssm_in.weight q4_0 +blk.17.ffn_down_exps.weight q5_1 +blk.17.ffn_down_shexp.weight q5_1 +blk.17.ffn_up_exps.weight q4_0 +blk.17.ffn_up_shexp.weight q4_0 +blk.18.ssm_in.weight q4_0 +blk.19.attn_k.weight q4_0 blk.19.attn_output.weight q5_K -blk.19.attn_v.weight q5_K -blk.20.ffn_down_exps.weight q5_K -blk.20.ffn_down_shexp.weight q5_K -blk.22.ffn_down_exps.weight q5_K -blk.22.ffn_down_shexp.weight q5_K -blk.24.ffn_down_exps.weight q5_K -blk.24.ffn_down_shexp.weight q5_K +blk.19.attn_q.weight q4_0 +blk.19.attn_v.weight q5_1 +blk.20.ffn_down_exps.weight q5_1 +blk.20.ffn_down_shexp.weight q5_1 +blk.20.ffn_up_exps.weight q4_0 +blk.20.ffn_up_shexp.weight q4_0 +blk.21.ssm_in.weight q4_0 +blk.22.ffn_down_exps.weight q5_1 +blk.22.ffn_down_shexp.weight q5_1 +blk.22.ffn_up_exps.weight q4_0 +blk.22.ffn_up_shexp.weight q4_0 +blk.23.ssm_in.weight q4_0 +blk.24.ffn_down_exps.weight q5_1 +blk.24.ffn_down_shexp.weight q5_1 +blk.24.ffn_up_exps.weight q4_0 +blk.24.ffn_up_shexp.weight q4_0 +blk.25.ssm_in.weight q4_0 +blk.26.attn_k.weight q4_0 blk.26.attn_output.weight q5_K -blk.26.attn_v.weight q5_K -blk.27.ffn_down_exps.weight q5_K -blk.27.ffn_down_shexp.weight q5_K -blk.29.ffn_down_exps.weight q5_K -blk.29.ffn_down_shexp.weight q5_K -blk.31.ffn_down_exps.weight q5_K -blk.31.ffn_down_shexp.weight q5_K +blk.26.attn_q.weight q4_0 +blk.26.attn_v.weight q5_1 +blk.27.ffn_down_exps.weight q5_1 +blk.27.ffn_down_shexp.weight q5_1 +blk.27.ffn_up_exps.weight q4_0 +blk.27.ffn_up_shexp.weight q4_0 +blk.28.ssm_in.weight q4_0 +blk.29.ffn_down_exps.weight q5_1 +blk.29.ffn_down_shexp.weight q5_1 +blk.29.ffn_up_exps.weight q4_0 +blk.29.ffn_up_shexp.weight q4_0 +blk.30.ssm_in.weight q4_0 +blk.31.ffn_down_exps.weight q5_1 +blk.31.ffn_down_shexp.weight q5_1 +blk.31.ffn_up_exps.weight q4_0 +blk.31.ffn_up_shexp.weight q4_0 +blk.32.ssm_in.weight q4_0 +blk.33.attn_k.weight q4_0 blk.33.attn_output.weight q5_K -blk.33.attn_v.weight q5_K -blk.34.ffn_down_exps.weight q5_K -blk.34.ffn_down_shexp.weight q5_K -blk.36.ffn_down_exps.weight q5_K -blk.36.ffn_down_shexp.weight q5_K -blk.38.ffn_down_exps.weight q5_K -blk.38.ffn_down_shexp.weight q5_K -blk.40.ffn_down_exps.weight q5_K -blk.40.ffn_down_shexp.weight q5_K +blk.33.attn_q.weight q4_0 +blk.33.attn_v.weight q5_1 +blk.34.ffn_down_exps.weight q5_1 +blk.34.ffn_down_shexp.weight q5_1 +blk.34.ffn_up_exps.weight q4_0 +blk.34.ffn_up_shexp.weight q4_0 +blk.35.ssm_in.weight q4_0 +blk.36.ffn_down_exps.weight q5_1 +blk.36.ffn_down_shexp.weight q5_1 +blk.36.ffn_up_exps.weight q4_0 +blk.36.ffn_up_shexp.weight q4_0 +blk.37.ssm_in.weight q4_0 +blk.38.ffn_down_exps.weight q5_1 +blk.38.ffn_down_shexp.weight q5_1 +blk.38.ffn_up_exps.weight q4_0 +blk.38.ffn_up_shexp.weight q4_0 +blk.39.ssm_in.weight q4_0 +blk.40.ffn_down_exps.weight q5_1 +blk.40.ffn_down_shexp.weight q5_1 +blk.40.ffn_up_exps.weight q4_0 +blk.40.ffn_up_shexp.weight q4_0 +blk.41.ssm_in.weight q4_0 +blk.42.attn_k.weight q4_0 blk.42.attn_output.weight q5_K -blk.42.attn_v.weight q5_K -blk.43.ffn_down_exps.weight q5_K -blk.43.ffn_down_shexp.weight q5_K -blk.45.ffn_down_exps.weight q5_K -blk.45.ffn_down_shexp.weight q5_K -blk.47.ffn_down_exps.weight q5_K -blk.47.ffn_down_shexp.weight q5_K -blk.49.ffn_down_exps.weight q5_K -blk.49.ffn_down_shexp.weight q5_K -blk.51.ffn_down_exps.weight q5_K -blk.51.ffn_down_shexp.weight q5_K +blk.42.attn_q.weight q4_0 +blk.42.attn_v.weight q5_1 +blk.43.ffn_down_exps.weight q5_1 +blk.43.ffn_down_shexp.weight q5_1 +blk.43.ffn_up_exps.weight q4_0 +blk.43.ffn_up_shexp.weight q4_0 +blk.44.ssm_in.weight q4_0 +blk.45.ffn_down_exps.weight q5_1 +blk.45.ffn_down_shexp.weight q5_1 +blk.45.ffn_up_exps.weight q4_0 +blk.45.ffn_up_shexp.weight q4_0 +blk.46.ssm_in.weight q4_0 +blk.47.ffn_down_exps.weight q5_1 +blk.47.ffn_down_shexp.weight q5_1 +blk.47.ffn_up_exps.weight q4_0 +blk.47.ffn_up_shexp.weight q4_0 +blk.48.ssm_in.weight q4_0 +blk.49.ffn_down_exps.weight q5_1 +blk.49.ffn_down_shexp.weight q5_1 +blk.49.ffn_up_exps.weight q4_0 +blk.49.ffn_up_shexp.weight q4_0 +blk.50.ssm_in.weight q4_0 +blk.51.ffn_down_exps.weight q5_1 +blk.51.ffn_down_shexp.weight q5_1 +blk.51.ffn_up_exps.weight q4_0 +blk.51.ffn_up_shexp.weight q4_0 [Q4_K_S] q4_K output.weight q8_0 -blk.1.ffn_down_exps.weight q5_K -blk.1.ffn_down_shexp.weight q5_K -blk.3.ffn_down_exps.weight q5_K -blk.3.ffn_down_shexp.weight q5_K -blk.5.attn_v.weight q5_K -blk.12.attn_v.weight q5_K -blk.19.attn_v.weight q5_K -blk.26.attn_v.weight q5_K +token_embd.weight q5_0 +blk.0.ssm_in.weight q5_0 +blk.1.ffn_down_exps.weight q5_1 +blk.1.ffn_down_shexp.weight q5_1 +blk.1.ffn_up_exps.weight q5_0 +blk.1.ffn_up_shexp.weight q5_0 +blk.2.ssm_in.weight q5_0 +blk.3.ffn_down_exps.weight q5_1 +blk.3.ffn_down_shexp.weight q5_1 +blk.3.ffn_up_exps.weight q5_0 +blk.3.ffn_up_shexp.weight q5_0 +blk.4.ssm_in.weight q5_0 +blk.5.attn_k.weight q5_0 +blk.5.attn_q.weight q5_0 +blk.5.attn_v.weight q5_1 +blk.6.ffn_down_exps.weight q5_0 +blk.6.ffn_down_shexp.weight q5_0 +blk.6.ffn_up_exps.weight q5_0 +blk.6.ffn_up_shexp.weight q5_0 +blk.7.ssm_in.weight q5_0 +blk.8.ffn_down_exps.weight q5_0 +blk.8.ffn_down_shexp.weight q5_0 +blk.8.ffn_up_exps.weight q5_0 +blk.8.ffn_up_shexp.weight q5_0 +blk.9.ssm_in.weight q5_0 +blk.10.ffn_down_exps.weight q5_0 +blk.10.ffn_down_shexp.weight q5_0 +blk.10.ffn_up_exps.weight q5_0 +blk.10.ffn_up_shexp.weight q5_0 +blk.11.ssm_in.weight q5_0 +blk.12.attn_k.weight q5_0 +blk.12.attn_q.weight q5_0 +blk.12.attn_v.weight q5_1 +blk.13.ffn_down_exps.weight q5_0 +blk.13.ffn_down_shexp.weight q5_0 +blk.13.ffn_up_exps.weight q5_0 +blk.13.ffn_up_shexp.weight q5_0 +blk.14.ssm_in.weight q5_0 +blk.15.ffn_down_exps.weight q5_0 +blk.15.ffn_down_shexp.weight q5_0 +blk.15.ffn_up_exps.weight q5_0 +blk.15.ffn_up_shexp.weight q5_0 +blk.16.ssm_in.weight q5_0 +blk.17.ffn_down_exps.weight q5_0 +blk.17.ffn_down_shexp.weight q5_0 +blk.17.ffn_up_exps.weight q5_0 +blk.17.ffn_up_shexp.weight q5_0 +blk.18.ssm_in.weight q5_0 +blk.19.attn_k.weight q5_0 +blk.19.attn_q.weight q5_0 +blk.19.attn_v.weight q5_1 +blk.20.ffn_down_exps.weight q5_0 +blk.20.ffn_down_shexp.weight q5_0 +blk.20.ffn_up_exps.weight q5_0 +blk.20.ffn_up_shexp.weight q5_0 +blk.21.ssm_in.weight q5_0 +blk.22.ffn_down_exps.weight q5_0 +blk.22.ffn_down_shexp.weight q5_0 +blk.22.ffn_up_exps.weight q5_0 +blk.22.ffn_up_shexp.weight q5_0 +blk.23.ssm_in.weight q5_0 +blk.24.ffn_down_exps.weight q5_0 +blk.24.ffn_down_shexp.weight q5_0 +blk.24.ffn_up_exps.weight q5_0 +blk.24.ffn_up_shexp.weight q5_0 +blk.25.ssm_in.weight q5_0 +blk.26.attn_k.weight q5_0 +blk.26.attn_q.weight q5_0 +blk.26.attn_v.weight q5_1 +blk.27.ffn_down_exps.weight q5_0 +blk.27.ffn_down_shexp.weight q5_0 +blk.27.ffn_up_exps.weight q5_0 +blk.27.ffn_up_shexp.weight q5_0 +blk.28.ssm_in.weight q5_0 +blk.29.ffn_down_exps.weight q5_0 +blk.29.ffn_down_shexp.weight q5_0 +blk.29.ffn_up_exps.weight q5_0 +blk.29.ffn_up_shexp.weight q5_0 +blk.30.ssm_in.weight q5_0 +blk.31.ffn_down_exps.weight q5_0 +blk.31.ffn_down_shexp.weight q5_0 +blk.31.ffn_up_exps.weight q5_0 +blk.31.ffn_up_shexp.weight q5_0 +blk.32.ssm_in.weight q5_0 +blk.33.attn_k.weight q5_0 +blk.33.attn_q.weight q5_0 +blk.33.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight q5_0 +blk.34.ffn_down_shexp.weight q5_0 +blk.34.ffn_up_exps.weight q5_0 +blk.34.ffn_up_shexp.weight q5_0 +blk.35.ssm_in.weight q5_0 +blk.36.ffn_down_exps.weight q5_0 +blk.36.ffn_down_shexp.weight q5_0 +blk.36.ffn_up_exps.weight q5_0 +blk.36.ffn_up_shexp.weight q5_0 +blk.37.ssm_in.weight q5_0 +blk.38.ffn_down_exps.weight q5_0 +blk.38.ffn_down_shexp.weight q5_0 +blk.38.ffn_up_exps.weight q5_0 +blk.38.ffn_up_shexp.weight q5_0 +blk.39.ssm_in.weight q5_0 +blk.40.ffn_down_exps.weight q5_0 +blk.40.ffn_down_shexp.weight q5_0 +blk.40.ffn_up_exps.weight q5_0 +blk.40.ffn_up_shexp.weight q5_0 +blk.41.ssm_in.weight q5_0 +blk.42.attn_k.weight q5_0 +blk.42.attn_q.weight q5_0 +blk.42.attn_v.weight q5_0 +blk.43.ffn_down_exps.weight q5_0 +blk.43.ffn_down_shexp.weight q5_0 +blk.43.ffn_up_exps.weight q5_0 +blk.43.ffn_up_shexp.weight q5_0 +blk.44.ssm_in.weight q5_0 +blk.45.ffn_down_exps.weight q5_0 +blk.45.ffn_down_shexp.weight q5_0 +blk.45.ffn_up_exps.weight q5_0 +blk.45.ffn_up_shexp.weight q5_0 +blk.46.ssm_in.weight q5_0 +blk.47.ffn_down_exps.weight q5_0 +blk.47.ffn_down_shexp.weight q5_0 +blk.47.ffn_up_exps.weight q5_0 +blk.47.ffn_up_shexp.weight q5_0 +blk.48.ssm_in.weight q5_0 +blk.49.ffn_down_exps.weight q5_0 +blk.49.ffn_down_shexp.weight q5_0 +blk.49.ffn_up_exps.weight q5_0 +blk.49.ffn_up_shexp.weight q5_0 +blk.50.ssm_in.weight q5_0 +blk.51.ffn_down_exps.weight q5_0 +blk.51.ffn_down_shexp.weight q5_0 +blk.51.ffn_up_exps.weight q5_0 +blk.51.ffn_up_shexp.weight q5_0 [Q4_K_M] q4_K output.weight q8_0 -blk.1.ffn_down_exps.weight q6_K -blk.1.ffn_down_shexp.weight q6_K -blk.3.ffn_down_exps.weight q6_K -blk.3.ffn_down_shexp.weight q6_K -blk.8.ffn_down_exps.weight q6_K -blk.8.ffn_down_shexp.weight q6_K -blk.17.ffn_down_exps.weight q6_K -blk.17.ffn_down_shexp.weight q6_K -blk.19.attn_v.weight q6_K -blk.20.ffn_down_exps.weight q6_K -blk.20.ffn_down_shexp.weight q6_K -blk.29.ffn_down_exps.weight q6_K -blk.29.ffn_down_shexp.weight q6_K -blk.38.ffn_down_exps.weight q6_K -blk.38.ffn_down_shexp.weight q6_K -blk.42.attn_v.weight q6_K -blk.45.ffn_down_exps.weight q6_K -blk.45.ffn_down_shexp.weight q6_K -blk.47.ffn_down_exps.weight q6_K -blk.47.ffn_down_shexp.weight q6_K -blk.49.ffn_down_exps.weight q6_K -blk.49.ffn_down_shexp.weight q6_K -blk.51.ffn_down_exps.weight q6_K -blk.51.ffn_down_shexp.weight q6_K +token_embd.weight q5_0 +blk.0.ssm_in.weight q5_0 +blk.1.ffn_down_exps.weight q8_0 +blk.1.ffn_down_shexp.weight q8_0 +blk.1.ffn_up_exps.weight q5_0 +blk.1.ffn_up_shexp.weight q5_0 +blk.2.ssm_in.weight q5_0 +blk.3.ffn_down_exps.weight q8_0 +blk.3.ffn_down_shexp.weight q8_0 +blk.3.ffn_up_exps.weight q5_0 +blk.3.ffn_up_shexp.weight q5_0 +blk.4.ssm_in.weight q5_0 +blk.5.attn_k.weight q5_0 +blk.5.attn_q.weight q5_0 +blk.5.attn_v.weight q5_0 +blk.6.ffn_down_exps.weight q5_0 +blk.6.ffn_down_shexp.weight q5_0 +blk.6.ffn_up_exps.weight q5_0 +blk.6.ffn_up_shexp.weight q5_0 +blk.7.ssm_in.weight q5_0 +blk.8.ffn_down_exps.weight q8_0 +blk.8.ffn_down_shexp.weight q8_0 +blk.8.ffn_up_exps.weight q5_0 +blk.8.ffn_up_shexp.weight q5_0 +blk.9.ssm_in.weight q5_0 +blk.10.ffn_down_exps.weight q5_0 +blk.10.ffn_down_shexp.weight q5_0 +blk.10.ffn_up_exps.weight q5_0 +blk.10.ffn_up_shexp.weight q5_0 +blk.11.ssm_in.weight q5_0 +blk.12.attn_k.weight q5_0 +blk.12.attn_q.weight q5_0 +blk.12.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight q5_0 +blk.13.ffn_down_shexp.weight q5_0 +blk.13.ffn_up_exps.weight q5_0 +blk.13.ffn_up_shexp.weight q5_0 +blk.14.ssm_in.weight q5_0 +blk.15.ffn_down_exps.weight q5_0 +blk.15.ffn_down_shexp.weight q5_0 +blk.15.ffn_up_exps.weight q5_0 +blk.15.ffn_up_shexp.weight q5_0 +blk.16.ssm_in.weight q5_0 +blk.17.ffn_down_exps.weight q8_0 +blk.17.ffn_down_shexp.weight q8_0 +blk.17.ffn_up_exps.weight q5_0 +blk.17.ffn_up_shexp.weight q5_0 +blk.18.ssm_in.weight q5_0 +blk.19.attn_k.weight q5_0 +blk.19.attn_q.weight q5_0 +blk.19.attn_v.weight q8_0 +blk.20.ffn_down_exps.weight q8_0 +blk.20.ffn_down_shexp.weight q8_0 +blk.20.ffn_up_exps.weight q5_0 +blk.20.ffn_up_shexp.weight q5_0 +blk.21.ssm_in.weight q5_0 +blk.22.ffn_down_exps.weight q5_0 +blk.22.ffn_down_shexp.weight q5_0 +blk.22.ffn_up_exps.weight q5_0 +blk.22.ffn_up_shexp.weight q5_0 +blk.23.ssm_in.weight q5_0 +blk.24.ffn_down_exps.weight q5_0 +blk.24.ffn_down_shexp.weight q5_0 +blk.24.ffn_up_exps.weight q5_0 +blk.24.ffn_up_shexp.weight q5_0 +blk.25.ssm_in.weight q5_0 +blk.26.attn_k.weight q5_0 +blk.26.attn_q.weight q5_0 +blk.26.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight q5_0 +blk.27.ffn_down_shexp.weight q5_0 +blk.27.ffn_up_exps.weight q5_0 +blk.27.ffn_up_shexp.weight q5_0 +blk.28.ssm_in.weight q5_0 +blk.29.ffn_down_exps.weight q8_0 +blk.29.ffn_down_shexp.weight q8_0 +blk.29.ffn_up_exps.weight q5_0 +blk.29.ffn_up_shexp.weight q5_0 +blk.30.ssm_in.weight q5_0 +blk.31.ffn_down_exps.weight q5_0 +blk.31.ffn_down_shexp.weight q5_0 +blk.31.ffn_up_exps.weight q5_0 +blk.31.ffn_up_shexp.weight q5_0 +blk.32.ssm_in.weight q5_0 +blk.33.attn_k.weight q5_0 +blk.33.attn_q.weight q5_0 +blk.33.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight q5_0 +blk.34.ffn_down_shexp.weight q5_0 +blk.34.ffn_up_exps.weight q5_0 +blk.34.ffn_up_shexp.weight q5_0 +blk.35.ssm_in.weight q5_0 +blk.36.ffn_down_exps.weight q5_0 +blk.36.ffn_down_shexp.weight q5_0 +blk.36.ffn_up_exps.weight q5_0 +blk.36.ffn_up_shexp.weight q5_0 +blk.37.ssm_in.weight q5_0 +blk.38.ffn_down_exps.weight q8_0 +blk.38.ffn_down_shexp.weight q8_0 +blk.38.ffn_up_exps.weight q5_0 +blk.38.ffn_up_shexp.weight q5_0 +blk.39.ssm_in.weight q5_0 +blk.40.ffn_down_exps.weight q5_0 +blk.40.ffn_down_shexp.weight q5_0 +blk.40.ffn_up_exps.weight q5_0 +blk.40.ffn_up_shexp.weight q5_0 +blk.41.ssm_in.weight q5_0 +blk.42.attn_k.weight q5_0 +blk.42.attn_q.weight q5_0 +blk.42.attn_v.weight q8_0 +blk.43.ffn_down_exps.weight q5_0 +blk.43.ffn_down_shexp.weight q5_0 +blk.43.ffn_up_exps.weight q5_0 +blk.43.ffn_up_shexp.weight q5_0 +blk.44.ssm_in.weight q5_0 +blk.45.ffn_down_exps.weight q8_0 +blk.45.ffn_down_shexp.weight q8_0 +blk.45.ffn_up_exps.weight q5_0 +blk.45.ffn_up_shexp.weight q5_0 +blk.46.ssm_in.weight q5_0 +blk.47.ffn_down_exps.weight q8_0 +blk.47.ffn_down_shexp.weight q8_0 +blk.47.ffn_up_exps.weight q5_0 +blk.47.ffn_up_shexp.weight q5_0 +blk.48.ssm_in.weight q5_0 +blk.49.ffn_down_exps.weight q8_0 +blk.49.ffn_down_shexp.weight q8_0 +blk.49.ffn_up_exps.weight q5_0 +blk.49.ffn_up_shexp.weight q5_0 +blk.50.ssm_in.weight q5_0 +blk.51.ffn_down_exps.weight q8_0 +blk.51.ffn_down_shexp.weight q8_0 +blk.51.ffn_up_exps.weight q5_0 +blk.51.ffn_up_shexp.weight q5_0 [Q5_K_S] q5_K output.weight q8_0 +token_embd.weight q5_1 +blk.0.ssm_in.weight q5_1 +blk.1.ffn_down_exps.weight q5_1 +blk.1.ffn_down_shexp.weight q5_1 +blk.1.ffn_up_exps.weight q5_1 +blk.1.ffn_up_shexp.weight q5_1 +blk.2.ssm_in.weight q5_1 +blk.3.ffn_down_exps.weight q5_1 +blk.3.ffn_down_shexp.weight q5_1 +blk.3.ffn_up_exps.weight q5_1 +blk.3.ffn_up_shexp.weight q5_1 +blk.4.ssm_in.weight q5_1 +blk.5.attn_k.weight q5_1 +blk.5.attn_q.weight q5_1 +blk.5.attn_v.weight q5_1 +blk.6.ffn_down_exps.weight q5_1 +blk.6.ffn_down_shexp.weight q5_1 +blk.6.ffn_up_exps.weight q5_1 +blk.6.ffn_up_shexp.weight q5_1 +blk.7.ssm_in.weight q5_1 +blk.8.ffn_down_exps.weight q5_1 +blk.8.ffn_down_shexp.weight q5_1 +blk.8.ffn_up_exps.weight q5_1 +blk.8.ffn_up_shexp.weight q5_1 +blk.9.ssm_in.weight q5_1 +blk.10.ffn_down_exps.weight q5_1 +blk.10.ffn_down_shexp.weight q5_1 +blk.10.ffn_up_exps.weight q5_1 +blk.10.ffn_up_shexp.weight q5_1 +blk.11.ssm_in.weight q5_1 +blk.12.attn_k.weight q5_1 +blk.12.attn_q.weight q5_1 +blk.12.attn_v.weight q5_1 +blk.13.ffn_down_exps.weight q5_1 +blk.13.ffn_down_shexp.weight q5_1 +blk.13.ffn_up_exps.weight q5_1 +blk.13.ffn_up_shexp.weight q5_1 +blk.14.ssm_in.weight q5_1 +blk.15.ffn_down_exps.weight q5_1 +blk.15.ffn_down_shexp.weight q5_1 +blk.15.ffn_up_exps.weight q5_1 +blk.15.ffn_up_shexp.weight q5_1 +blk.16.ssm_in.weight q5_1 +blk.17.ffn_down_exps.weight q5_1 +blk.17.ffn_down_shexp.weight q5_1 +blk.17.ffn_up_exps.weight q5_1 +blk.17.ffn_up_shexp.weight q5_1 +blk.18.ssm_in.weight q5_1 +blk.19.attn_k.weight q5_1 +blk.19.attn_q.weight q5_1 +blk.19.attn_v.weight q5_1 +blk.20.ffn_down_exps.weight q5_1 +blk.20.ffn_down_shexp.weight q5_1 +blk.20.ffn_up_exps.weight q5_1 +blk.20.ffn_up_shexp.weight q5_1 +blk.21.ssm_in.weight q5_1 +blk.22.ffn_down_exps.weight q5_1 +blk.22.ffn_down_shexp.weight q5_1 +blk.22.ffn_up_exps.weight q5_1 +blk.22.ffn_up_shexp.weight q5_1 +blk.23.ssm_in.weight q5_1 +blk.24.ffn_down_exps.weight q5_1 +blk.24.ffn_down_shexp.weight q5_1 +blk.24.ffn_up_exps.weight q5_1 +blk.24.ffn_up_shexp.weight q5_1 +blk.25.ssm_in.weight q5_1 +blk.26.attn_k.weight q5_1 +blk.26.attn_q.weight q5_1 +blk.26.attn_v.weight q5_1 +blk.27.ffn_down_exps.weight q5_1 +blk.27.ffn_down_shexp.weight q5_1 +blk.27.ffn_up_exps.weight q5_1 +blk.27.ffn_up_shexp.weight q5_1 +blk.28.ssm_in.weight q5_1 +blk.29.ffn_down_exps.weight q5_1 +blk.29.ffn_down_shexp.weight q5_1 +blk.29.ffn_up_exps.weight q5_1 +blk.29.ffn_up_shexp.weight q5_1 +blk.30.ssm_in.weight q5_1 +blk.31.ffn_down_exps.weight q5_1 +blk.31.ffn_down_shexp.weight q5_1 +blk.31.ffn_up_exps.weight q5_1 +blk.31.ffn_up_shexp.weight q5_1 +blk.32.ssm_in.weight q5_1 +blk.33.attn_k.weight q5_1 +blk.33.attn_q.weight q5_1 +blk.33.attn_v.weight q5_1 +blk.34.ffn_down_exps.weight q5_1 +blk.34.ffn_down_shexp.weight q5_1 +blk.34.ffn_up_exps.weight q5_1 +blk.34.ffn_up_shexp.weight q5_1 +blk.35.ssm_in.weight q5_1 +blk.36.ffn_down_exps.weight q5_1 +blk.36.ffn_down_shexp.weight q5_1 +blk.36.ffn_up_exps.weight q5_1 +blk.36.ffn_up_shexp.weight q5_1 +blk.37.ssm_in.weight q5_1 +blk.38.ffn_down_exps.weight q5_1 +blk.38.ffn_down_shexp.weight q5_1 +blk.38.ffn_up_exps.weight q5_1 +blk.38.ffn_up_shexp.weight q5_1 +blk.39.ssm_in.weight q5_1 +blk.40.ffn_down_exps.weight q5_1 +blk.40.ffn_down_shexp.weight q5_1 +blk.40.ffn_up_exps.weight q5_1 +blk.40.ffn_up_shexp.weight q5_1 +blk.41.ssm_in.weight q5_1 +blk.42.attn_k.weight q5_1 +blk.42.attn_q.weight q5_1 +blk.42.attn_v.weight q5_1 +blk.43.ffn_down_exps.weight q5_1 +blk.43.ffn_down_shexp.weight q5_1 +blk.43.ffn_up_exps.weight q5_1 +blk.43.ffn_up_shexp.weight q5_1 +blk.44.ssm_in.weight q5_1 +blk.45.ffn_down_exps.weight q5_1 +blk.45.ffn_down_shexp.weight q5_1 +blk.45.ffn_up_exps.weight q5_1 +blk.45.ffn_up_shexp.weight q5_1 +blk.46.ssm_in.weight q5_1 +blk.47.ffn_down_exps.weight q5_1 +blk.47.ffn_down_shexp.weight q5_1 +blk.47.ffn_up_exps.weight q5_1 +blk.47.ffn_up_shexp.weight q5_1 +blk.48.ssm_in.weight q5_1 +blk.49.ffn_down_exps.weight q5_1 +blk.49.ffn_down_shexp.weight q5_1 +blk.49.ffn_up_exps.weight q5_1 +blk.49.ffn_up_shexp.weight q5_1 +blk.50.ssm_in.weight q5_1 +blk.51.ffn_down_exps.weight q5_1 +blk.51.ffn_down_shexp.weight q5_1 +blk.51.ffn_up_exps.weight q5_1 +blk.51.ffn_up_shexp.weight q5_1 [Q5_K_M] q5_K output.weight q8_0 -blk.1.ffn_down_exps.weight q6_K -blk.1.ffn_down_shexp.weight q6_K -blk.3.ffn_down_exps.weight q6_K -blk.3.ffn_down_shexp.weight q6_K -blk.8.ffn_down_exps.weight q6_K -blk.8.ffn_down_shexp.weight q6_K -blk.17.ffn_down_exps.weight q6_K -blk.17.ffn_down_shexp.weight q6_K -blk.19.attn_v.weight q6_K -blk.20.ffn_down_exps.weight q6_K -blk.20.ffn_down_shexp.weight q6_K -blk.29.ffn_down_exps.weight q6_K -blk.29.ffn_down_shexp.weight q6_K -blk.38.ffn_down_exps.weight q6_K -blk.38.ffn_down_shexp.weight q6_K -blk.42.attn_v.weight q6_K -blk.45.ffn_down_exps.weight q6_K -blk.45.ffn_down_shexp.weight q6_K -blk.47.ffn_down_exps.weight q6_K -blk.47.ffn_down_shexp.weight q6_K -blk.49.ffn_down_exps.weight q6_K -blk.49.ffn_down_shexp.weight q6_K -blk.51.ffn_down_exps.weight q6_K -blk.51.ffn_down_shexp.weight q6_K +token_embd.weight q5_1 +blk.0.ssm_in.weight q5_1 +blk.1.ffn_down_exps.weight q8_0 +blk.1.ffn_down_shexp.weight q8_0 +blk.1.ffn_up_exps.weight q5_1 +blk.1.ffn_up_shexp.weight q5_1 +blk.2.ssm_in.weight q5_1 +blk.3.ffn_down_exps.weight q8_0 +blk.3.ffn_down_shexp.weight q8_0 +blk.3.ffn_up_exps.weight q5_1 +blk.3.ffn_up_shexp.weight q5_1 +blk.4.ssm_in.weight q5_1 +blk.5.attn_k.weight q5_1 +blk.5.attn_q.weight q5_1 +blk.5.attn_v.weight q5_1 +blk.6.ffn_down_exps.weight q5_1 +blk.6.ffn_down_shexp.weight q5_1 +blk.6.ffn_up_exps.weight q5_1 +blk.6.ffn_up_shexp.weight q5_1 +blk.7.ssm_in.weight q5_1 +blk.8.ffn_down_exps.weight q8_0 +blk.8.ffn_down_shexp.weight q8_0 +blk.8.ffn_up_exps.weight q5_1 +blk.8.ffn_up_shexp.weight q5_1 +blk.9.ssm_in.weight q5_1 +blk.10.ffn_down_exps.weight q5_1 +blk.10.ffn_down_shexp.weight q5_1 +blk.10.ffn_up_exps.weight q5_1 +blk.10.ffn_up_shexp.weight q5_1 +blk.11.ssm_in.weight q5_1 +blk.12.attn_k.weight q5_1 +blk.12.attn_q.weight q5_1 +blk.12.attn_v.weight q5_1 +blk.13.ffn_down_exps.weight q5_1 +blk.13.ffn_down_shexp.weight q5_1 +blk.13.ffn_up_exps.weight q5_1 +blk.13.ffn_up_shexp.weight q5_1 +blk.14.ssm_in.weight q5_1 +blk.15.ffn_down_exps.weight q5_1 +blk.15.ffn_down_shexp.weight q5_1 +blk.15.ffn_up_exps.weight q5_1 +blk.15.ffn_up_shexp.weight q5_1 +blk.16.ssm_in.weight q5_1 +blk.17.ffn_down_exps.weight q8_0 +blk.17.ffn_down_shexp.weight q8_0 +blk.17.ffn_up_exps.weight q5_1 +blk.17.ffn_up_shexp.weight q5_1 +blk.18.ssm_in.weight q5_1 +blk.19.attn_k.weight q5_1 +blk.19.attn_q.weight q5_1 +blk.19.attn_v.weight q8_0 +blk.20.ffn_down_exps.weight q8_0 +blk.20.ffn_down_shexp.weight q8_0 +blk.20.ffn_up_exps.weight q5_1 +blk.20.ffn_up_shexp.weight q5_1 +blk.21.ssm_in.weight q5_1 +blk.22.ffn_down_exps.weight q5_1 +blk.22.ffn_down_shexp.weight q5_1 +blk.22.ffn_up_exps.weight q5_1 +blk.22.ffn_up_shexp.weight q5_1 +blk.23.ssm_in.weight q5_1 +blk.24.ffn_down_exps.weight q5_1 +blk.24.ffn_down_shexp.weight q5_1 +blk.24.ffn_up_exps.weight q5_1 +blk.24.ffn_up_shexp.weight q5_1 +blk.25.ssm_in.weight q5_1 +blk.26.attn_k.weight q5_1 +blk.26.attn_q.weight q5_1 +blk.26.attn_v.weight q5_1 +blk.27.ffn_down_exps.weight q5_1 +blk.27.ffn_down_shexp.weight q5_1 +blk.27.ffn_up_exps.weight q5_1 +blk.27.ffn_up_shexp.weight q5_1 +blk.28.ssm_in.weight q5_1 +blk.29.ffn_down_exps.weight q8_0 +blk.29.ffn_down_shexp.weight q8_0 +blk.29.ffn_up_exps.weight q5_1 +blk.29.ffn_up_shexp.weight q5_1 +blk.30.ssm_in.weight q5_1 +blk.31.ffn_down_exps.weight q5_1 +blk.31.ffn_down_shexp.weight q5_1 +blk.31.ffn_up_exps.weight q5_1 +blk.31.ffn_up_shexp.weight q5_1 +blk.32.ssm_in.weight q5_1 +blk.33.attn_k.weight q5_1 +blk.33.attn_q.weight q5_1 +blk.33.attn_v.weight q5_1 +blk.34.ffn_down_exps.weight q5_1 +blk.34.ffn_down_shexp.weight q5_1 +blk.34.ffn_up_exps.weight q5_1 +blk.34.ffn_up_shexp.weight q5_1 +blk.35.ssm_in.weight q5_1 +blk.36.ffn_down_exps.weight q5_1 +blk.36.ffn_down_shexp.weight q5_1 +blk.36.ffn_up_exps.weight q5_1 +blk.36.ffn_up_shexp.weight q5_1 +blk.37.ssm_in.weight q5_1 +blk.38.ffn_down_exps.weight q8_0 +blk.38.ffn_down_shexp.weight q8_0 +blk.38.ffn_up_exps.weight q5_1 +blk.38.ffn_up_shexp.weight q5_1 +blk.39.ssm_in.weight q5_1 +blk.40.ffn_down_exps.weight q5_1 +blk.40.ffn_down_shexp.weight q5_1 +blk.40.ffn_up_exps.weight q5_1 +blk.40.ffn_up_shexp.weight q5_1 +blk.41.ssm_in.weight q5_1 +blk.42.attn_k.weight q5_1 +blk.42.attn_q.weight q5_1 +blk.42.attn_v.weight q8_0 +blk.43.ffn_down_exps.weight q5_1 +blk.43.ffn_down_shexp.weight q5_1 +blk.43.ffn_up_exps.weight q5_1 +blk.43.ffn_up_shexp.weight q5_1 +blk.44.ssm_in.weight q5_1 +blk.45.ffn_down_exps.weight q8_0 +blk.45.ffn_down_shexp.weight q8_0 +blk.45.ffn_up_exps.weight q5_1 +blk.45.ffn_up_shexp.weight q5_1 +blk.46.ssm_in.weight q5_1 +blk.47.ffn_down_exps.weight q8_0 +blk.47.ffn_down_shexp.weight q8_0 +blk.47.ffn_up_exps.weight q5_1 +blk.47.ffn_up_shexp.weight q5_1 +blk.48.ssm_in.weight q5_1 +blk.49.ffn_down_exps.weight q8_0 +blk.49.ffn_down_shexp.weight q8_0 +blk.49.ffn_up_exps.weight q5_1 +blk.49.ffn_up_shexp.weight q5_1 +blk.50.ssm_in.weight q5_1 +blk.51.ffn_down_exps.weight q8_0 +blk.51.ffn_down_shexp.weight q8_0 +blk.51.ffn_up_exps.weight q5_1 +blk.51.ffn_up_shexp.weight q5_1 [Q6_K] q6_K output.weight q8_0 +token_embd.weight q8_0 +blk.0.ssm_in.weight q8_0 +blk.1.ffn_down_exps.weight q8_0 +blk.1.ffn_down_shexp.weight q8_0 +blk.1.ffn_up_exps.weight q8_0 +blk.1.ffn_up_shexp.weight q8_0 +blk.2.ssm_in.weight q8_0 +blk.3.ffn_down_exps.weight q8_0 +blk.3.ffn_down_shexp.weight q8_0 +blk.3.ffn_up_exps.weight q8_0 +blk.3.ffn_up_shexp.weight q8_0 +blk.4.ssm_in.weight q8_0 +blk.5.attn_k.weight q8_0 +blk.5.attn_q.weight q8_0 +blk.5.attn_v.weight q8_0 +blk.6.ffn_down_exps.weight q8_0 +blk.6.ffn_down_shexp.weight q8_0 +blk.6.ffn_up_exps.weight q8_0 +blk.6.ffn_up_shexp.weight q8_0 +blk.7.ssm_in.weight q8_0 +blk.8.ffn_down_exps.weight q8_0 +blk.8.ffn_down_shexp.weight q8_0 +blk.8.ffn_up_exps.weight q8_0 +blk.8.ffn_up_shexp.weight q8_0 +blk.9.ssm_in.weight q8_0 +blk.10.ffn_down_exps.weight q8_0 +blk.10.ffn_down_shexp.weight q8_0 +blk.10.ffn_up_exps.weight q8_0 +blk.10.ffn_up_shexp.weight q8_0 +blk.11.ssm_in.weight q8_0 +blk.12.attn_k.weight q8_0 +blk.12.attn_q.weight q8_0 +blk.12.attn_v.weight q8_0 +blk.13.ffn_down_exps.weight q8_0 +blk.13.ffn_down_shexp.weight q8_0 +blk.13.ffn_up_exps.weight q8_0 +blk.13.ffn_up_shexp.weight q8_0 +blk.14.ssm_in.weight q8_0 +blk.15.ffn_down_exps.weight q8_0 +blk.15.ffn_down_shexp.weight q8_0 +blk.15.ffn_up_exps.weight q8_0 +blk.15.ffn_up_shexp.weight q8_0 +blk.16.ssm_in.weight q8_0 +blk.17.ffn_down_exps.weight q8_0 +blk.17.ffn_down_shexp.weight q8_0 +blk.17.ffn_up_exps.weight q8_0 +blk.17.ffn_up_shexp.weight q8_0 +blk.18.ssm_in.weight q8_0 +blk.19.attn_k.weight q8_0 +blk.19.attn_q.weight q8_0 +blk.19.attn_v.weight q8_0 +blk.20.ffn_down_exps.weight q8_0 +blk.20.ffn_down_shexp.weight q8_0 +blk.20.ffn_up_exps.weight q8_0 +blk.20.ffn_up_shexp.weight q8_0 +blk.21.ssm_in.weight q8_0 +blk.22.ffn_down_exps.weight q8_0 +blk.22.ffn_down_shexp.weight q8_0 +blk.22.ffn_up_exps.weight q8_0 +blk.22.ffn_up_shexp.weight q8_0 +blk.23.ssm_in.weight q8_0 +blk.24.ffn_down_exps.weight q8_0 +blk.24.ffn_down_shexp.weight q8_0 +blk.24.ffn_up_exps.weight q8_0 +blk.24.ffn_up_shexp.weight q8_0 +blk.25.ssm_in.weight q8_0 +blk.26.attn_k.weight q8_0 +blk.26.attn_q.weight q8_0 +blk.26.attn_v.weight q8_0 +blk.27.ffn_down_exps.weight q8_0 +blk.27.ffn_down_shexp.weight q8_0 +blk.27.ffn_up_exps.weight q8_0 +blk.27.ffn_up_shexp.weight q8_0 +blk.28.ssm_in.weight q8_0 +blk.29.ffn_down_exps.weight q8_0 +blk.29.ffn_down_shexp.weight q8_0 +blk.29.ffn_up_exps.weight q8_0 +blk.29.ffn_up_shexp.weight q8_0 +blk.30.ssm_in.weight q8_0 +blk.31.ffn_down_exps.weight q8_0 +blk.31.ffn_down_shexp.weight q8_0 +blk.31.ffn_up_exps.weight q8_0 +blk.31.ffn_up_shexp.weight q8_0 +blk.32.ssm_in.weight q8_0 +blk.33.attn_k.weight q8_0 +blk.33.attn_q.weight q8_0 +blk.33.attn_v.weight q8_0 +blk.34.ffn_down_exps.weight q8_0 +blk.34.ffn_down_shexp.weight q8_0 +blk.34.ffn_up_exps.weight q8_0 +blk.34.ffn_up_shexp.weight q8_0 +blk.35.ssm_in.weight q8_0 +blk.36.ffn_down_exps.weight q8_0 +blk.36.ffn_down_shexp.weight q8_0 +blk.36.ffn_up_exps.weight q8_0 +blk.36.ffn_up_shexp.weight q8_0 +blk.37.ssm_in.weight q8_0 +blk.38.ffn_down_exps.weight q8_0 +blk.38.ffn_down_shexp.weight q8_0 +blk.38.ffn_up_exps.weight q8_0 +blk.38.ffn_up_shexp.weight q8_0 +blk.39.ssm_in.weight q8_0 +blk.40.ffn_down_exps.weight q8_0 +blk.40.ffn_down_shexp.weight q8_0 +blk.40.ffn_up_exps.weight q8_0 +blk.40.ffn_up_shexp.weight q8_0 +blk.41.ssm_in.weight q8_0 +blk.42.attn_k.weight q8_0 +blk.42.attn_q.weight q8_0 +blk.42.attn_v.weight q8_0 +blk.43.ffn_down_exps.weight q8_0 +blk.43.ffn_down_shexp.weight q8_0 +blk.43.ffn_up_exps.weight q8_0 +blk.43.ffn_up_shexp.weight q8_0 +blk.44.ssm_in.weight q8_0 +blk.45.ffn_down_exps.weight q8_0 +blk.45.ffn_down_shexp.weight q8_0 +blk.45.ffn_up_exps.weight q8_0 +blk.45.ffn_up_shexp.weight q8_0 +blk.46.ssm_in.weight q8_0 +blk.47.ffn_down_exps.weight q8_0 +blk.47.ffn_down_shexp.weight q8_0 +blk.47.ffn_up_exps.weight q8_0 +blk.47.ffn_up_shexp.weight q8_0 +blk.48.ssm_in.weight q8_0 +blk.49.ffn_down_exps.weight q8_0 +blk.49.ffn_down_shexp.weight q8_0 +blk.49.ffn_up_exps.weight q8_0 +blk.49.ffn_up_shexp.weight q8_0 +blk.50.ssm_in.weight q8_0 +blk.51.ffn_down_exps.weight q8_0 +blk.51.ffn_down_shexp.weight q8_0 +blk.51.ffn_up_exps.weight q8_0 +blk.51.ffn_up_shexp.weight q8_0 [IQ2_XXS] iq2_xxs output.weight q8_0 -token_embd.weight q2_K -blk.1.ffn_down_exps.weight q2_K -blk.1.ffn_down_shexp.weight q2_K -blk.3.ffn_down_exps.weight q2_K -blk.3.ffn_down_shexp.weight q2_K -blk.5.attn_v.weight q4_K -blk.6.ffn_down_exps.weight q2_K -blk.6.ffn_down_shexp.weight q2_K -blk.12.attn_v.weight q4_K -blk.19.attn_v.weight q4_K -blk.26.attn_v.weight q4_K -blk.33.attn_v.weight q4_K -blk.42.attn_v.weight q4_K +token_embd.weight q4_0 +blk.0.ssm_in.weight iq4_nl +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_down_shexp.weight q4_0 +blk.1.ffn_up_exps.weight iq4_nl +blk.1.ffn_up_shexp.weight iq4_nl +blk.2.ssm_in.weight iq4_nl +blk.3.ffn_down_exps.weight q4_0 +blk.3.ffn_down_shexp.weight q4_0 +blk.3.ffn_up_exps.weight iq4_nl +blk.3.ffn_up_shexp.weight iq4_nl +blk.4.ssm_in.weight iq4_nl +blk.5.attn_k.weight iq4_nl +blk.5.attn_q.weight iq4_nl +blk.5.attn_v.weight q5_0 +blk.6.ffn_down_exps.weight q4_0 +blk.6.ffn_down_shexp.weight q4_0 +blk.6.ffn_up_exps.weight iq4_nl +blk.6.ffn_up_shexp.weight iq4_nl +blk.7.ssm_in.weight iq4_nl +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_down_shexp.weight iq4_nl +blk.8.ffn_up_exps.weight iq4_nl +blk.8.ffn_up_shexp.weight iq4_nl +blk.9.ssm_in.weight iq4_nl +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_down_shexp.weight iq4_nl +blk.10.ffn_up_exps.weight iq4_nl +blk.10.ffn_up_shexp.weight iq4_nl +blk.11.ssm_in.weight iq4_nl +blk.12.attn_k.weight iq4_nl +blk.12.attn_q.weight iq4_nl +blk.12.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_down_shexp.weight iq4_nl +blk.13.ffn_up_exps.weight iq4_nl +blk.13.ffn_up_shexp.weight iq4_nl +blk.14.ssm_in.weight iq4_nl +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_down_shexp.weight iq4_nl +blk.15.ffn_up_exps.weight iq4_nl +blk.15.ffn_up_shexp.weight iq4_nl +blk.16.ssm_in.weight iq4_nl +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_down_shexp.weight iq4_nl +blk.17.ffn_up_exps.weight iq4_nl +blk.17.ffn_up_shexp.weight iq4_nl +blk.18.ssm_in.weight iq4_nl +blk.19.attn_k.weight iq4_nl +blk.19.attn_q.weight iq4_nl +blk.19.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_down_shexp.weight iq4_nl +blk.20.ffn_up_exps.weight iq4_nl +blk.20.ffn_up_shexp.weight iq4_nl +blk.21.ssm_in.weight iq4_nl +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_down_shexp.weight iq4_nl +blk.22.ffn_up_exps.weight iq4_nl +blk.22.ffn_up_shexp.weight iq4_nl +blk.23.ssm_in.weight iq4_nl +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_down_shexp.weight iq4_nl +blk.24.ffn_up_exps.weight iq4_nl +blk.24.ffn_up_shexp.weight iq4_nl +blk.25.ssm_in.weight iq4_nl +blk.26.attn_k.weight iq4_nl +blk.26.attn_q.weight iq4_nl +blk.26.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_down_shexp.weight iq4_nl +blk.27.ffn_up_exps.weight iq4_nl +blk.27.ffn_up_shexp.weight iq4_nl +blk.28.ssm_in.weight iq4_nl +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_down_shexp.weight iq4_nl +blk.29.ffn_up_exps.weight iq4_nl +blk.29.ffn_up_shexp.weight iq4_nl +blk.30.ssm_in.weight iq4_nl +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_down_shexp.weight iq4_nl +blk.31.ffn_up_exps.weight iq4_nl +blk.31.ffn_up_shexp.weight iq4_nl +blk.32.ssm_in.weight iq4_nl +blk.33.attn_k.weight iq4_nl +blk.33.attn_q.weight iq4_nl +blk.33.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_down_shexp.weight iq4_nl +blk.34.ffn_up_exps.weight iq4_nl +blk.34.ffn_up_shexp.weight iq4_nl +blk.35.ssm_in.weight iq4_nl +blk.36.ffn_down_exps.weight iq4_nl +blk.36.ffn_down_shexp.weight iq4_nl +blk.36.ffn_up_exps.weight iq4_nl +blk.36.ffn_up_shexp.weight iq4_nl +blk.37.ssm_in.weight iq4_nl +blk.38.ffn_down_exps.weight iq4_nl +blk.38.ffn_down_shexp.weight iq4_nl +blk.38.ffn_up_exps.weight iq4_nl +blk.38.ffn_up_shexp.weight iq4_nl +blk.39.ssm_in.weight iq4_nl +blk.40.ffn_down_exps.weight iq4_nl +blk.40.ffn_down_shexp.weight iq4_nl +blk.40.ffn_up_exps.weight iq4_nl +blk.40.ffn_up_shexp.weight iq4_nl +blk.41.ssm_in.weight iq4_nl +blk.42.attn_k.weight iq4_nl +blk.42.attn_q.weight iq4_nl +blk.42.attn_v.weight q5_0 +blk.43.ffn_down_exps.weight iq4_nl +blk.43.ffn_down_shexp.weight iq4_nl +blk.43.ffn_up_exps.weight iq4_nl +blk.43.ffn_up_shexp.weight iq4_nl +blk.44.ssm_in.weight iq4_nl +blk.45.ffn_down_exps.weight iq4_nl +blk.45.ffn_down_shexp.weight iq4_nl +blk.45.ffn_up_exps.weight iq4_nl +blk.45.ffn_up_shexp.weight iq4_nl +blk.46.ssm_in.weight iq4_nl +blk.47.ffn_down_exps.weight iq4_nl +blk.47.ffn_down_shexp.weight iq4_nl +blk.47.ffn_up_exps.weight iq4_nl +blk.47.ffn_up_shexp.weight iq4_nl +blk.48.ssm_in.weight iq4_nl +blk.49.ffn_down_exps.weight iq4_nl +blk.49.ffn_down_shexp.weight iq4_nl +blk.49.ffn_up_exps.weight iq4_nl +blk.49.ffn_up_shexp.weight iq4_nl +blk.50.ssm_in.weight iq4_nl +blk.51.ffn_down_exps.weight iq4_nl +blk.51.ffn_down_shexp.weight iq4_nl +blk.51.ffn_up_exps.weight iq4_nl +blk.51.ffn_up_shexp.weight iq4_nl [IQ2_XS] iq2_xs output.weight q8_0 -token_embd.weight q2_K -blk.1.ffn_down_exps.weight q2_K -blk.1.ffn_down_shexp.weight q2_K -blk.3.ffn_down_exps.weight q2_K -blk.3.ffn_down_shexp.weight q2_K -blk.5.attn_v.weight q4_K -blk.6.ffn_down_exps.weight q2_K -blk.6.ffn_down_shexp.weight q2_K -blk.12.attn_v.weight q4_K -blk.19.attn_v.weight q4_K -blk.26.attn_v.weight q4_K -blk.33.attn_v.weight q4_K -blk.42.attn_v.weight q4_K +token_embd.weight q4_0 +blk.0.ssm_in.weight iq4_nl +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_down_shexp.weight q4_0 +blk.1.ffn_up_exps.weight iq4_nl +blk.1.ffn_up_shexp.weight iq4_nl +blk.2.ssm_in.weight iq4_nl +blk.3.ffn_down_exps.weight q4_0 +blk.3.ffn_down_shexp.weight q4_0 +blk.3.ffn_up_exps.weight iq4_nl +blk.3.ffn_up_shexp.weight iq4_nl +blk.4.ssm_in.weight iq4_nl +blk.5.attn_k.weight iq4_nl +blk.5.attn_q.weight iq4_nl +blk.5.attn_v.weight q5_0 +blk.6.ffn_down_exps.weight q4_0 +blk.6.ffn_down_shexp.weight q4_0 +blk.6.ffn_up_exps.weight iq4_nl +blk.6.ffn_up_shexp.weight iq4_nl +blk.7.ssm_in.weight iq4_nl +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_down_shexp.weight iq4_nl +blk.8.ffn_up_exps.weight iq4_nl +blk.8.ffn_up_shexp.weight iq4_nl +blk.9.ssm_in.weight iq4_nl +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_down_shexp.weight iq4_nl +blk.10.ffn_up_exps.weight iq4_nl +blk.10.ffn_up_shexp.weight iq4_nl +blk.11.ssm_in.weight iq4_nl +blk.12.attn_k.weight iq4_nl +blk.12.attn_q.weight iq4_nl +blk.12.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_down_shexp.weight iq4_nl +blk.13.ffn_up_exps.weight iq4_nl +blk.13.ffn_up_shexp.weight iq4_nl +blk.14.ssm_in.weight iq4_nl +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_down_shexp.weight iq4_nl +blk.15.ffn_up_exps.weight iq4_nl +blk.15.ffn_up_shexp.weight iq4_nl +blk.16.ssm_in.weight iq4_nl +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_down_shexp.weight iq4_nl +blk.17.ffn_up_exps.weight iq4_nl +blk.17.ffn_up_shexp.weight iq4_nl +blk.18.ssm_in.weight iq4_nl +blk.19.attn_k.weight iq4_nl +blk.19.attn_q.weight iq4_nl +blk.19.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_down_shexp.weight iq4_nl +blk.20.ffn_up_exps.weight iq4_nl +blk.20.ffn_up_shexp.weight iq4_nl +blk.21.ssm_in.weight iq4_nl +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_down_shexp.weight iq4_nl +blk.22.ffn_up_exps.weight iq4_nl +blk.22.ffn_up_shexp.weight iq4_nl +blk.23.ssm_in.weight iq4_nl +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_down_shexp.weight iq4_nl +blk.24.ffn_up_exps.weight iq4_nl +blk.24.ffn_up_shexp.weight iq4_nl +blk.25.ssm_in.weight iq4_nl +blk.26.attn_k.weight iq4_nl +blk.26.attn_q.weight iq4_nl +blk.26.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_down_shexp.weight iq4_nl +blk.27.ffn_up_exps.weight iq4_nl +blk.27.ffn_up_shexp.weight iq4_nl +blk.28.ssm_in.weight iq4_nl +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_down_shexp.weight iq4_nl +blk.29.ffn_up_exps.weight iq4_nl +blk.29.ffn_up_shexp.weight iq4_nl +blk.30.ssm_in.weight iq4_nl +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_down_shexp.weight iq4_nl +blk.31.ffn_up_exps.weight iq4_nl +blk.31.ffn_up_shexp.weight iq4_nl +blk.32.ssm_in.weight iq4_nl +blk.33.attn_k.weight iq4_nl +blk.33.attn_q.weight iq4_nl +blk.33.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_down_shexp.weight iq4_nl +blk.34.ffn_up_exps.weight iq4_nl +blk.34.ffn_up_shexp.weight iq4_nl +blk.35.ssm_in.weight iq4_nl +blk.36.ffn_down_exps.weight iq4_nl +blk.36.ffn_down_shexp.weight iq4_nl +blk.36.ffn_up_exps.weight iq4_nl +blk.36.ffn_up_shexp.weight iq4_nl +blk.37.ssm_in.weight iq4_nl +blk.38.ffn_down_exps.weight iq4_nl +blk.38.ffn_down_shexp.weight iq4_nl +blk.38.ffn_up_exps.weight iq4_nl +blk.38.ffn_up_shexp.weight iq4_nl +blk.39.ssm_in.weight iq4_nl +blk.40.ffn_down_exps.weight iq4_nl +blk.40.ffn_down_shexp.weight iq4_nl +blk.40.ffn_up_exps.weight iq4_nl +blk.40.ffn_up_shexp.weight iq4_nl +blk.41.ssm_in.weight iq4_nl +blk.42.attn_k.weight iq4_nl +blk.42.attn_q.weight iq4_nl +blk.42.attn_v.weight q5_0 +blk.43.ffn_down_exps.weight iq4_nl +blk.43.ffn_down_shexp.weight iq4_nl +blk.43.ffn_up_exps.weight iq4_nl +blk.43.ffn_up_shexp.weight iq4_nl +blk.44.ssm_in.weight iq4_nl +blk.45.ffn_down_exps.weight iq4_nl +blk.45.ffn_down_shexp.weight iq4_nl +blk.45.ffn_up_exps.weight iq4_nl +blk.45.ffn_up_shexp.weight iq4_nl +blk.46.ssm_in.weight iq4_nl +blk.47.ffn_down_exps.weight iq4_nl +blk.47.ffn_down_shexp.weight iq4_nl +blk.47.ffn_up_exps.weight iq4_nl +blk.47.ffn_up_shexp.weight iq4_nl +blk.48.ssm_in.weight iq4_nl +blk.49.ffn_down_exps.weight iq4_nl +blk.49.ffn_down_shexp.weight iq4_nl +blk.49.ffn_up_exps.weight iq4_nl +blk.49.ffn_up_shexp.weight iq4_nl +blk.50.ssm_in.weight iq4_nl +blk.51.ffn_down_exps.weight iq4_nl +blk.51.ffn_down_shexp.weight iq4_nl +blk.51.ffn_up_exps.weight iq4_nl +blk.51.ffn_up_shexp.weight iq4_nl [Q2_K_S] q2_K output.weight q8_0 -blk.1.ffn_down_exps.weight q4_K -blk.1.ffn_down_shexp.weight q4_K -blk.3.ffn_down_exps.weight q4_K -blk.3.ffn_down_shexp.weight q4_K +token_embd.weight q4_0 +blk.0.ssm_in.weight q4_0 +blk.1.ffn_down_exps.weight q5_0 +blk.1.ffn_down_shexp.weight q5_0 +blk.1.ffn_up_exps.weight q4_0 +blk.1.ffn_up_shexp.weight q4_0 +blk.2.ssm_in.weight q4_0 +blk.3.ffn_down_exps.weight q5_0 +blk.3.ffn_down_shexp.weight q5_0 +blk.3.ffn_up_exps.weight q4_0 +blk.3.ffn_up_shexp.weight q4_0 +blk.4.ssm_in.weight q4_0 +blk.5.attn_k.weight q4_0 +blk.5.attn_q.weight q4_0 +blk.5.attn_v.weight q4_0 +blk.6.ffn_down_exps.weight q4_0 +blk.6.ffn_down_shexp.weight q4_0 +blk.6.ffn_up_exps.weight q4_0 +blk.6.ffn_up_shexp.weight q4_0 +blk.7.ssm_in.weight q4_0 +blk.8.ffn_down_exps.weight q4_0 +blk.8.ffn_down_shexp.weight q4_0 +blk.8.ffn_up_exps.weight q4_0 +blk.8.ffn_up_shexp.weight q4_0 +blk.9.ssm_in.weight q4_0 +blk.10.ffn_down_exps.weight q4_0 +blk.10.ffn_down_shexp.weight q4_0 +blk.10.ffn_up_exps.weight q4_0 +blk.10.ffn_up_shexp.weight q4_0 +blk.11.ssm_in.weight q4_0 +blk.12.attn_k.weight q4_0 +blk.12.attn_q.weight q4_0 +blk.12.attn_v.weight q4_0 +blk.13.ffn_down_exps.weight q4_0 +blk.13.ffn_down_shexp.weight q4_0 +blk.13.ffn_up_exps.weight q4_0 +blk.13.ffn_up_shexp.weight q4_0 +blk.14.ssm_in.weight q4_0 +blk.15.ffn_down_exps.weight q4_0 +blk.15.ffn_down_shexp.weight q4_0 +blk.15.ffn_up_exps.weight q4_0 +blk.15.ffn_up_shexp.weight q4_0 +blk.16.ssm_in.weight q4_0 +blk.17.ffn_down_exps.weight q4_0 +blk.17.ffn_down_shexp.weight q4_0 +blk.17.ffn_up_exps.weight q4_0 +blk.17.ffn_up_shexp.weight q4_0 +blk.18.ssm_in.weight q4_0 +blk.19.attn_k.weight q4_0 +blk.19.attn_q.weight q4_0 +blk.19.attn_v.weight q4_0 +blk.20.ffn_down_exps.weight q4_0 +blk.20.ffn_down_shexp.weight q4_0 +blk.20.ffn_up_exps.weight q4_0 +blk.20.ffn_up_shexp.weight q4_0 +blk.21.ssm_in.weight q4_0 +blk.22.ffn_down_exps.weight q4_0 +blk.22.ffn_down_shexp.weight q4_0 +blk.22.ffn_up_exps.weight q4_0 +blk.22.ffn_up_shexp.weight q4_0 +blk.23.ssm_in.weight q4_0 +blk.24.ffn_down_exps.weight q4_0 +blk.24.ffn_down_shexp.weight q4_0 +blk.24.ffn_up_exps.weight q4_0 +blk.24.ffn_up_shexp.weight q4_0 +blk.25.ssm_in.weight q4_0 +blk.26.attn_k.weight q4_0 +blk.26.attn_q.weight q4_0 +blk.26.attn_v.weight q4_0 +blk.27.ffn_down_exps.weight q4_0 +blk.27.ffn_down_shexp.weight q4_0 +blk.27.ffn_up_exps.weight q4_0 +blk.27.ffn_up_shexp.weight q4_0 +blk.28.ssm_in.weight q4_0 +blk.29.ffn_down_exps.weight q4_0 +blk.29.ffn_down_shexp.weight q4_0 +blk.29.ffn_up_exps.weight q4_0 +blk.29.ffn_up_shexp.weight q4_0 +blk.30.ssm_in.weight q4_0 +blk.31.ffn_down_exps.weight q4_0 +blk.31.ffn_down_shexp.weight q4_0 +blk.31.ffn_up_exps.weight q4_0 +blk.31.ffn_up_shexp.weight q4_0 +blk.32.ssm_in.weight q4_0 +blk.33.attn_k.weight q4_0 +blk.33.attn_q.weight q4_0 +blk.33.attn_v.weight q4_0 +blk.34.ffn_down_exps.weight q4_0 +blk.34.ffn_down_shexp.weight q4_0 +blk.34.ffn_up_exps.weight q4_0 +blk.34.ffn_up_shexp.weight q4_0 +blk.35.ssm_in.weight q4_0 +blk.36.ffn_down_exps.weight q4_0 +blk.36.ffn_down_shexp.weight q4_0 +blk.36.ffn_up_exps.weight q4_0 +blk.36.ffn_up_shexp.weight q4_0 +blk.37.ssm_in.weight q4_0 +blk.38.ffn_down_exps.weight q4_0 +blk.38.ffn_down_shexp.weight q4_0 +blk.38.ffn_up_exps.weight q4_0 +blk.38.ffn_up_shexp.weight q4_0 +blk.39.ssm_in.weight q4_0 +blk.40.ffn_down_exps.weight q4_0 +blk.40.ffn_down_shexp.weight q4_0 +blk.40.ffn_up_exps.weight q4_0 +blk.40.ffn_up_shexp.weight q4_0 +blk.41.ssm_in.weight q4_0 +blk.42.attn_k.weight q4_0 +blk.42.attn_q.weight q4_0 +blk.42.attn_v.weight q4_0 +blk.43.ffn_down_exps.weight q4_0 +blk.43.ffn_down_shexp.weight q4_0 +blk.43.ffn_up_exps.weight q4_0 +blk.43.ffn_up_shexp.weight q4_0 +blk.44.ssm_in.weight q4_0 +blk.45.ffn_down_exps.weight q4_0 +blk.45.ffn_down_shexp.weight q4_0 +blk.45.ffn_up_exps.weight q4_0 +blk.45.ffn_up_shexp.weight q4_0 +blk.46.ssm_in.weight q4_0 +blk.47.ffn_down_exps.weight q4_0 +blk.47.ffn_down_shexp.weight q4_0 +blk.47.ffn_up_exps.weight q4_0 +blk.47.ffn_up_shexp.weight q4_0 +blk.48.ssm_in.weight q4_0 +blk.49.ffn_down_exps.weight q4_0 +blk.49.ffn_down_shexp.weight q4_0 +blk.49.ffn_up_exps.weight q4_0 +blk.49.ffn_up_shexp.weight q4_0 +blk.50.ssm_in.weight q4_0 +blk.51.ffn_down_exps.weight q4_0 +blk.51.ffn_down_shexp.weight q4_0 +blk.51.ffn_up_exps.weight q4_0 +blk.51.ffn_up_shexp.weight q4_0 [IQ3_XS] iq3_s output.weight q8_0 -blk.5.attn_k.weight iq3_xxs -blk.5.attn_q.weight iq3_xxs -blk.6.ffn_up_exps.weight iq3_xxs -blk.6.ffn_up_shexp.weight iq3_xxs -blk.8.ffn_up_exps.weight iq3_xxs -blk.8.ffn_up_shexp.weight iq3_xxs -blk.10.ffn_up_exps.weight iq3_xxs -blk.10.ffn_up_shexp.weight iq3_xxs -blk.12.attn_k.weight iq3_xxs -blk.12.attn_q.weight iq3_xxs -blk.13.ffn_up_exps.weight iq3_xxs -blk.13.ffn_up_shexp.weight iq3_xxs -blk.15.ffn_up_exps.weight iq3_xxs -blk.15.ffn_up_shexp.weight iq3_xxs -blk.17.ffn_up_exps.weight iq3_xxs -blk.17.ffn_up_shexp.weight iq3_xxs -blk.19.attn_k.weight iq3_xxs -blk.19.attn_q.weight iq3_xxs -blk.20.ffn_up_exps.weight iq3_xxs -blk.20.ffn_up_shexp.weight iq3_xxs -blk.22.ffn_up_exps.weight iq3_xxs -blk.22.ffn_up_shexp.weight iq3_xxs -blk.24.ffn_up_exps.weight iq3_xxs -blk.24.ffn_up_shexp.weight iq3_xxs -blk.26.attn_k.weight iq3_xxs -blk.26.attn_q.weight iq3_xxs -blk.27.ffn_up_exps.weight iq3_xxs -blk.27.ffn_up_shexp.weight iq3_xxs -blk.29.ffn_up_exps.weight iq3_xxs -blk.29.ffn_up_shexp.weight iq3_xxs -blk.31.ffn_up_exps.weight iq3_xxs -blk.31.ffn_up_shexp.weight iq3_xxs -blk.33.attn_k.weight iq3_xxs -blk.33.attn_q.weight iq3_xxs -blk.34.ffn_up_exps.weight iq3_xxs -blk.34.ffn_up_shexp.weight iq3_xxs -blk.36.ffn_up_exps.weight iq3_xxs -blk.36.ffn_up_shexp.weight iq3_xxs -blk.38.ffn_up_exps.weight iq3_xxs -blk.38.ffn_up_shexp.weight iq3_xxs -blk.40.ffn_up_exps.weight iq3_xxs -blk.40.ffn_up_shexp.weight iq3_xxs -blk.42.attn_k.weight iq3_xxs -blk.42.attn_q.weight iq3_xxs -blk.43.ffn_up_exps.weight iq3_xxs -blk.43.ffn_up_shexp.weight iq3_xxs +token_embd.weight iq4_nl +blk.0.ssm_in.weight iq4_nl +blk.1.ffn_down_exps.weight iq4_nl +blk.1.ffn_down_shexp.weight iq4_nl +blk.1.ffn_up_exps.weight iq4_nl +blk.1.ffn_up_shexp.weight iq4_nl +blk.2.ssm_in.weight iq4_nl +blk.3.ffn_down_exps.weight iq4_nl +blk.3.ffn_down_shexp.weight iq4_nl +blk.3.ffn_up_exps.weight iq4_nl +blk.3.ffn_up_shexp.weight iq4_nl +blk.4.ssm_in.weight iq4_nl +blk.5.attn_k.weight iq4_nl +blk.5.attn_q.weight iq4_nl +blk.5.attn_v.weight iq4_nl +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_down_shexp.weight iq4_nl +blk.6.ffn_up_exps.weight iq4_nl +blk.6.ffn_up_shexp.weight iq4_nl +blk.7.ssm_in.weight iq4_nl +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_down_shexp.weight iq4_nl +blk.8.ffn_up_exps.weight iq4_nl +blk.8.ffn_up_shexp.weight iq4_nl +blk.9.ssm_in.weight iq4_nl +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_down_shexp.weight iq4_nl +blk.10.ffn_up_exps.weight iq4_nl +blk.10.ffn_up_shexp.weight iq4_nl +blk.11.ssm_in.weight iq4_nl +blk.12.attn_k.weight iq4_nl +blk.12.attn_q.weight iq4_nl +blk.12.attn_v.weight iq4_nl +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_down_shexp.weight iq4_nl +blk.13.ffn_up_exps.weight iq4_nl +blk.13.ffn_up_shexp.weight iq4_nl +blk.14.ssm_in.weight iq4_nl +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_down_shexp.weight iq4_nl +blk.15.ffn_up_exps.weight iq4_nl +blk.15.ffn_up_shexp.weight iq4_nl +blk.16.ssm_in.weight iq4_nl +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_down_shexp.weight iq4_nl +blk.17.ffn_up_exps.weight iq4_nl +blk.17.ffn_up_shexp.weight iq4_nl +blk.18.ssm_in.weight iq4_nl +blk.19.attn_k.weight iq4_nl +blk.19.attn_q.weight iq4_nl +blk.19.attn_v.weight iq4_nl +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_down_shexp.weight iq4_nl +blk.20.ffn_up_exps.weight iq4_nl +blk.20.ffn_up_shexp.weight iq4_nl +blk.21.ssm_in.weight iq4_nl +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_down_shexp.weight iq4_nl +blk.22.ffn_up_exps.weight iq4_nl +blk.22.ffn_up_shexp.weight iq4_nl +blk.23.ssm_in.weight iq4_nl +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_down_shexp.weight iq4_nl +blk.24.ffn_up_exps.weight iq4_nl +blk.24.ffn_up_shexp.weight iq4_nl +blk.25.ssm_in.weight iq4_nl +blk.26.attn_k.weight iq4_nl +blk.26.attn_q.weight iq4_nl +blk.26.attn_v.weight iq4_nl +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_down_shexp.weight iq4_nl +blk.27.ffn_up_exps.weight iq4_nl +blk.27.ffn_up_shexp.weight iq4_nl +blk.28.ssm_in.weight iq4_nl +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_down_shexp.weight iq4_nl +blk.29.ffn_up_exps.weight iq4_nl +blk.29.ffn_up_shexp.weight iq4_nl +blk.30.ssm_in.weight iq4_nl +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_down_shexp.weight iq4_nl +blk.31.ffn_up_exps.weight iq4_nl +blk.31.ffn_up_shexp.weight iq4_nl +blk.32.ssm_in.weight iq4_nl +blk.33.attn_k.weight iq4_nl +blk.33.attn_q.weight iq4_nl +blk.33.attn_v.weight iq4_nl +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_down_shexp.weight iq4_nl +blk.34.ffn_up_exps.weight iq4_nl +blk.34.ffn_up_shexp.weight iq4_nl +blk.35.ssm_in.weight iq4_nl +blk.36.ffn_down_exps.weight iq4_nl +blk.36.ffn_down_shexp.weight iq4_nl +blk.36.ffn_up_exps.weight iq4_nl +blk.36.ffn_up_shexp.weight iq4_nl +blk.37.ssm_in.weight iq4_nl +blk.38.ffn_down_exps.weight iq4_nl +blk.38.ffn_down_shexp.weight iq4_nl +blk.38.ffn_up_exps.weight iq4_nl +blk.38.ffn_up_shexp.weight iq4_nl +blk.39.ssm_in.weight iq4_nl +blk.40.ffn_down_exps.weight iq4_nl +blk.40.ffn_down_shexp.weight iq4_nl +blk.40.ffn_up_exps.weight iq4_nl +blk.40.ffn_up_shexp.weight iq4_nl +blk.41.ssm_in.weight iq4_nl +blk.42.attn_k.weight iq4_nl +blk.42.attn_q.weight iq4_nl +blk.42.attn_v.weight iq4_nl +blk.43.ffn_down_exps.weight iq4_nl +blk.43.ffn_down_shexp.weight iq4_nl +blk.43.ffn_up_exps.weight iq4_nl +blk.43.ffn_up_shexp.weight iq4_nl +blk.44.ssm_in.weight iq4_nl +blk.45.ffn_down_exps.weight iq4_nl +blk.45.ffn_down_shexp.weight iq4_nl +blk.45.ffn_up_exps.weight iq4_nl +blk.45.ffn_up_shexp.weight iq4_nl +blk.46.ssm_in.weight iq4_nl +blk.47.ffn_down_exps.weight iq4_nl +blk.47.ffn_down_shexp.weight iq4_nl +blk.47.ffn_up_exps.weight iq4_nl +blk.47.ffn_up_shexp.weight iq4_nl +blk.48.ssm_in.weight iq4_nl +blk.49.ffn_down_exps.weight iq4_nl +blk.49.ffn_down_shexp.weight iq4_nl +blk.49.ffn_up_exps.weight iq4_nl +blk.49.ffn_up_shexp.weight iq4_nl +blk.50.ssm_in.weight iq4_nl +blk.51.ffn_down_exps.weight iq4_nl +blk.51.ffn_down_shexp.weight iq4_nl +blk.51.ffn_up_exps.weight iq4_nl +blk.51.ffn_up_shexp.weight iq4_nl [IQ3_XXS] iq3_xxs output.weight q8_0 -token_embd.weight iq3_s -blk.1.ffn_down_exps.weight q4_K -blk.1.ffn_down_shexp.weight q4_K -blk.3.ffn_down_exps.weight q4_K -blk.3.ffn_down_shexp.weight q4_K -blk.5.attn_k.weight iq2_s +token_embd.weight iq4_nl +blk.0.ssm_in.weight iq4_nl +blk.1.ffn_down_exps.weight q5_0 +blk.1.ffn_down_shexp.weight q5_0 +blk.1.ffn_up_exps.weight iq4_nl +blk.1.ffn_up_shexp.weight iq4_nl +blk.2.ssm_in.weight iq4_nl +blk.3.ffn_down_exps.weight q5_0 +blk.3.ffn_down_shexp.weight q5_0 +blk.3.ffn_up_exps.weight iq4_nl +blk.3.ffn_up_shexp.weight iq4_nl +blk.4.ssm_in.weight iq4_nl +blk.5.attn_k.weight iq4_nl blk.5.attn_output.weight iq3_s -blk.5.attn_q.weight iq2_s -blk.5.attn_v.weight iq3_s -blk.6.ffn_down_exps.weight q3_K -blk.6.ffn_down_shexp.weight q3_K -blk.8.ffn_down_exps.weight q3_K -blk.8.ffn_down_shexp.weight q3_K -blk.10.ffn_down_exps.weight q3_K -blk.10.ffn_down_shexp.weight q3_K -blk.12.attn_k.weight iq2_s +blk.5.attn_q.weight iq4_nl +blk.5.attn_v.weight iq4_nl +blk.6.ffn_down_exps.weight q4_0 +blk.6.ffn_down_shexp.weight q4_0 +blk.6.ffn_up_exps.weight iq4_nl +blk.6.ffn_up_shexp.weight iq4_nl +blk.7.ssm_in.weight iq4_nl +blk.8.ffn_down_exps.weight q4_0 +blk.8.ffn_down_shexp.weight q4_0 +blk.8.ffn_up_exps.weight iq4_nl +blk.8.ffn_up_shexp.weight iq4_nl +blk.9.ssm_in.weight iq4_nl +blk.10.ffn_down_exps.weight q4_0 +blk.10.ffn_down_shexp.weight q4_0 +blk.10.ffn_up_exps.weight iq4_nl +blk.10.ffn_up_shexp.weight iq4_nl +blk.11.ssm_in.weight iq4_nl +blk.12.attn_k.weight iq4_nl blk.12.attn_output.weight iq3_s -blk.12.attn_q.weight iq2_s -blk.12.attn_v.weight iq3_s -blk.13.ffn_down_exps.weight q3_K -blk.13.ffn_down_shexp.weight q3_K -blk.15.ffn_down_exps.weight q3_K -blk.15.ffn_down_shexp.weight q3_K -blk.17.ffn_down_exps.weight q3_K -blk.17.ffn_down_shexp.weight q3_K -blk.19.attn_k.weight iq2_s +blk.12.attn_q.weight iq4_nl +blk.12.attn_v.weight iq4_nl +blk.13.ffn_down_exps.weight q4_0 +blk.13.ffn_down_shexp.weight q4_0 +blk.13.ffn_up_exps.weight iq4_nl +blk.13.ffn_up_shexp.weight iq4_nl +blk.14.ssm_in.weight iq4_nl +blk.15.ffn_down_exps.weight q4_0 +blk.15.ffn_down_shexp.weight q4_0 +blk.15.ffn_up_exps.weight iq4_nl +blk.15.ffn_up_shexp.weight iq4_nl +blk.16.ssm_in.weight iq4_nl +blk.17.ffn_down_exps.weight q4_0 +blk.17.ffn_down_shexp.weight q4_0 +blk.17.ffn_up_exps.weight iq4_nl +blk.17.ffn_up_shexp.weight iq4_nl +blk.18.ssm_in.weight iq4_nl +blk.19.attn_k.weight iq4_nl blk.19.attn_output.weight iq3_s -blk.19.attn_q.weight iq2_s -blk.19.attn_v.weight iq3_s -blk.20.ffn_down_exps.weight q3_K -blk.20.ffn_down_shexp.weight q3_K -blk.22.ffn_down_exps.weight q3_K -blk.22.ffn_down_shexp.weight q3_K -blk.24.ffn_down_exps.weight q3_K -blk.24.ffn_down_shexp.weight q3_K -blk.26.attn_k.weight iq2_s +blk.19.attn_q.weight iq4_nl +blk.19.attn_v.weight iq4_nl +blk.20.ffn_down_exps.weight q4_0 +blk.20.ffn_down_shexp.weight q4_0 +blk.20.ffn_up_exps.weight iq4_nl +blk.20.ffn_up_shexp.weight iq4_nl +blk.21.ssm_in.weight iq4_nl +blk.22.ffn_down_exps.weight q4_0 +blk.22.ffn_down_shexp.weight q4_0 +blk.22.ffn_up_exps.weight iq4_nl +blk.22.ffn_up_shexp.weight iq4_nl +blk.23.ssm_in.weight iq4_nl +blk.24.ffn_down_exps.weight q4_0 +blk.24.ffn_down_shexp.weight q4_0 +blk.24.ffn_up_exps.weight iq4_nl +blk.24.ffn_up_shexp.weight iq4_nl +blk.25.ssm_in.weight iq4_nl +blk.26.attn_k.weight iq4_nl blk.26.attn_output.weight iq3_s -blk.26.attn_q.weight iq2_s -blk.26.attn_v.weight iq3_s -blk.27.ffn_down_exps.weight q3_K -blk.27.ffn_down_shexp.weight q3_K -blk.29.ffn_down_exps.weight q3_K -blk.29.ffn_down_shexp.weight q3_K -blk.31.ffn_down_exps.weight q3_K -blk.31.ffn_down_shexp.weight q3_K -blk.33.attn_k.weight iq2_s +blk.26.attn_q.weight iq4_nl +blk.26.attn_v.weight iq4_nl +blk.27.ffn_down_exps.weight q4_0 +blk.27.ffn_down_shexp.weight q4_0 +blk.27.ffn_up_exps.weight iq4_nl +blk.27.ffn_up_shexp.weight iq4_nl +blk.28.ssm_in.weight iq4_nl +blk.29.ffn_down_exps.weight q4_0 +blk.29.ffn_down_shexp.weight q4_0 +blk.29.ffn_up_exps.weight iq4_nl +blk.29.ffn_up_shexp.weight iq4_nl +blk.30.ssm_in.weight iq4_nl +blk.31.ffn_down_exps.weight q4_0 +blk.31.ffn_down_shexp.weight q4_0 +blk.31.ffn_up_exps.weight iq4_nl +blk.31.ffn_up_shexp.weight iq4_nl +blk.32.ssm_in.weight iq4_nl +blk.33.attn_k.weight iq4_nl blk.33.attn_output.weight iq3_s -blk.33.attn_q.weight iq2_s -blk.33.attn_v.weight iq3_s -blk.34.ffn_down_exps.weight q3_K -blk.34.ffn_down_shexp.weight q3_K -blk.36.ffn_down_exps.weight q3_K -blk.36.ffn_down_shexp.weight q3_K -blk.38.ffn_down_exps.weight q3_K -blk.38.ffn_down_shexp.weight q3_K -blk.40.ffn_down_exps.weight q3_K -blk.40.ffn_down_shexp.weight q3_K -blk.42.attn_k.weight iq2_s +blk.33.attn_q.weight iq4_nl +blk.33.attn_v.weight iq4_nl +blk.34.ffn_down_exps.weight q4_0 +blk.34.ffn_down_shexp.weight q4_0 +blk.34.ffn_up_exps.weight iq4_nl +blk.34.ffn_up_shexp.weight iq4_nl +blk.35.ssm_in.weight iq4_nl +blk.36.ffn_down_exps.weight q4_0 +blk.36.ffn_down_shexp.weight q4_0 +blk.36.ffn_up_exps.weight iq4_nl +blk.36.ffn_up_shexp.weight iq4_nl +blk.37.ssm_in.weight iq4_nl +blk.38.ffn_down_exps.weight q4_0 +blk.38.ffn_down_shexp.weight q4_0 +blk.38.ffn_up_exps.weight iq4_nl +blk.38.ffn_up_shexp.weight iq4_nl +blk.39.ssm_in.weight iq4_nl +blk.40.ffn_down_exps.weight q4_0 +blk.40.ffn_down_shexp.weight q4_0 +blk.40.ffn_up_exps.weight iq4_nl +blk.40.ffn_up_shexp.weight iq4_nl +blk.41.ssm_in.weight iq4_nl +blk.42.attn_k.weight iq4_nl blk.42.attn_output.weight iq3_s -blk.42.attn_q.weight iq2_s -blk.42.attn_v.weight iq3_s -blk.43.ffn_down_exps.weight q3_K -blk.43.ffn_down_shexp.weight q3_K -blk.45.ffn_down_exps.weight q3_K -blk.45.ffn_down_shexp.weight q3_K -blk.47.ffn_down_exps.weight q3_K -blk.47.ffn_down_shexp.weight q3_K -blk.49.ffn_down_exps.weight q3_K -blk.49.ffn_down_shexp.weight q3_K -blk.51.ffn_down_exps.weight q3_K -blk.51.ffn_down_shexp.weight q3_K +blk.42.attn_q.weight iq4_nl +blk.42.attn_v.weight iq4_nl +blk.43.ffn_down_exps.weight q4_0 +blk.43.ffn_down_shexp.weight q4_0 +blk.43.ffn_up_exps.weight iq4_nl +blk.43.ffn_up_shexp.weight iq4_nl +blk.44.ssm_in.weight iq4_nl +blk.45.ffn_down_exps.weight q4_0 +blk.45.ffn_down_shexp.weight q4_0 +blk.45.ffn_up_exps.weight iq4_nl +blk.45.ffn_up_shexp.weight iq4_nl +blk.46.ssm_in.weight iq4_nl +blk.47.ffn_down_exps.weight q4_0 +blk.47.ffn_down_shexp.weight q4_0 +blk.47.ffn_up_exps.weight iq4_nl +blk.47.ffn_up_shexp.weight iq4_nl +blk.48.ssm_in.weight iq4_nl +blk.49.ffn_down_exps.weight q4_0 +blk.49.ffn_down_shexp.weight q4_0 +blk.49.ffn_up_exps.weight iq4_nl +blk.49.ffn_up_shexp.weight iq4_nl +blk.50.ssm_in.weight iq4_nl +blk.51.ffn_down_exps.weight q4_0 +blk.51.ffn_down_shexp.weight q4_0 +blk.51.ffn_up_exps.weight iq4_nl +blk.51.ffn_up_shexp.weight iq4_nl [IQ1_S] iq1_s output.weight q8_0 -token_embd.weight q2_K -blk.1.ffn_down_exps.weight q2_K -blk.1.ffn_down_shexp.weight q2_K -blk.3.ffn_down_exps.weight q2_K -blk.3.ffn_down_shexp.weight q2_K +token_embd.weight q4_0 +blk.0.ssm_in.weight iq4_nl +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_down_shexp.weight q4_0 +blk.1.ffn_up_exps.weight iq4_nl +blk.1.ffn_up_shexp.weight iq4_nl +blk.2.ssm_in.weight iq4_nl +blk.3.ffn_down_exps.weight q4_0 +blk.3.ffn_down_shexp.weight q4_0 +blk.3.ffn_up_exps.weight iq4_nl +blk.3.ffn_up_shexp.weight iq4_nl +blk.4.ssm_in.weight iq4_nl +blk.5.attn_k.weight iq4_nl blk.5.attn_output.weight iq2_xxs -blk.5.attn_v.weight q4_K -blk.6.ffn_down_exps.weight q2_K -blk.6.ffn_down_shexp.weight q2_K +blk.5.attn_q.weight iq4_nl +blk.5.attn_v.weight q5_0 +blk.6.ffn_down_exps.weight q4_0 +blk.6.ffn_down_shexp.weight q4_0 +blk.6.ffn_up_exps.weight iq4_nl +blk.6.ffn_up_shexp.weight iq4_nl +blk.7.ssm_in.weight iq4_nl +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_down_shexp.weight iq4_nl +blk.8.ffn_up_exps.weight iq4_nl +blk.8.ffn_up_shexp.weight iq4_nl +blk.9.ssm_in.weight iq4_nl +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_down_shexp.weight iq4_nl +blk.10.ffn_up_exps.weight iq4_nl +blk.10.ffn_up_shexp.weight iq4_nl +blk.11.ssm_in.weight iq4_nl +blk.12.attn_k.weight iq4_nl blk.12.attn_output.weight iq2_xxs -blk.12.attn_v.weight q4_K +blk.12.attn_q.weight iq4_nl +blk.12.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_down_shexp.weight iq4_nl +blk.13.ffn_up_exps.weight iq4_nl +blk.13.ffn_up_shexp.weight iq4_nl +blk.14.ssm_in.weight iq4_nl +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_down_shexp.weight iq4_nl +blk.15.ffn_up_exps.weight iq4_nl +blk.15.ffn_up_shexp.weight iq4_nl +blk.16.ssm_in.weight iq4_nl +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_down_shexp.weight iq4_nl +blk.17.ffn_up_exps.weight iq4_nl +blk.17.ffn_up_shexp.weight iq4_nl +blk.18.ssm_in.weight iq4_nl +blk.19.attn_k.weight iq4_nl blk.19.attn_output.weight iq2_xxs -blk.19.attn_v.weight q4_K +blk.19.attn_q.weight iq4_nl +blk.19.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_down_shexp.weight iq4_nl +blk.20.ffn_up_exps.weight iq4_nl +blk.20.ffn_up_shexp.weight iq4_nl +blk.21.ssm_in.weight iq4_nl +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_down_shexp.weight iq4_nl +blk.22.ffn_up_exps.weight iq4_nl +blk.22.ffn_up_shexp.weight iq4_nl +blk.23.ssm_in.weight iq4_nl +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_down_shexp.weight iq4_nl +blk.24.ffn_up_exps.weight iq4_nl +blk.24.ffn_up_shexp.weight iq4_nl +blk.25.ssm_in.weight iq4_nl +blk.26.attn_k.weight iq4_nl blk.26.attn_output.weight iq2_xxs -blk.26.attn_v.weight q4_K +blk.26.attn_q.weight iq4_nl +blk.26.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_down_shexp.weight iq4_nl +blk.27.ffn_up_exps.weight iq4_nl +blk.27.ffn_up_shexp.weight iq4_nl +blk.28.ssm_in.weight iq4_nl +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_down_shexp.weight iq4_nl +blk.29.ffn_up_exps.weight iq4_nl +blk.29.ffn_up_shexp.weight iq4_nl +blk.30.ssm_in.weight iq4_nl +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_down_shexp.weight iq4_nl +blk.31.ffn_up_exps.weight iq4_nl +blk.31.ffn_up_shexp.weight iq4_nl +blk.32.ssm_in.weight iq4_nl +blk.33.attn_k.weight iq4_nl blk.33.attn_output.weight iq2_xxs -blk.33.attn_v.weight q4_K +blk.33.attn_q.weight iq4_nl +blk.33.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_down_shexp.weight iq4_nl +blk.34.ffn_up_exps.weight iq4_nl +blk.34.ffn_up_shexp.weight iq4_nl +blk.35.ssm_in.weight iq4_nl +blk.36.ffn_down_exps.weight iq4_nl +blk.36.ffn_down_shexp.weight iq4_nl +blk.36.ffn_up_exps.weight iq4_nl +blk.36.ffn_up_shexp.weight iq4_nl +blk.37.ssm_in.weight iq4_nl +blk.38.ffn_down_exps.weight iq4_nl +blk.38.ffn_down_shexp.weight iq4_nl +blk.38.ffn_up_exps.weight iq4_nl +blk.38.ffn_up_shexp.weight iq4_nl +blk.39.ssm_in.weight iq4_nl +blk.40.ffn_down_exps.weight iq4_nl +blk.40.ffn_down_shexp.weight iq4_nl +blk.40.ffn_up_exps.weight iq4_nl +blk.40.ffn_up_shexp.weight iq4_nl +blk.41.ssm_in.weight iq4_nl +blk.42.attn_k.weight iq4_nl blk.42.attn_output.weight iq2_xxs -blk.42.attn_v.weight q4_K +blk.42.attn_q.weight iq4_nl +blk.42.attn_v.weight q5_0 +blk.43.ffn_down_exps.weight iq4_nl +blk.43.ffn_down_shexp.weight iq4_nl +blk.43.ffn_up_exps.weight iq4_nl +blk.43.ffn_up_shexp.weight iq4_nl +blk.44.ssm_in.weight iq4_nl +blk.45.ffn_down_exps.weight iq4_nl +blk.45.ffn_down_shexp.weight iq4_nl +blk.45.ffn_up_exps.weight iq4_nl +blk.45.ffn_up_shexp.weight iq4_nl +blk.46.ssm_in.weight iq4_nl +blk.47.ffn_down_exps.weight iq4_nl +blk.47.ffn_down_shexp.weight iq4_nl +blk.47.ffn_up_exps.weight iq4_nl +blk.47.ffn_up_shexp.weight iq4_nl +blk.48.ssm_in.weight iq4_nl +blk.49.ffn_down_exps.weight iq4_nl +blk.49.ffn_down_shexp.weight iq4_nl +blk.49.ffn_up_exps.weight iq4_nl +blk.49.ffn_up_shexp.weight iq4_nl +blk.50.ssm_in.weight iq4_nl +blk.51.ffn_down_exps.weight iq4_nl +blk.51.ffn_down_shexp.weight iq4_nl +blk.51.ffn_up_exps.weight iq4_nl +blk.51.ffn_up_shexp.weight iq4_nl [IQ4_NL] iq4_nl -output.weight q6_K -blk.1.ffn_down_exps.weight q5_K -blk.1.ffn_down_shexp.weight q5_K -blk.3.ffn_down_exps.weight q5_K -blk.3.ffn_down_shexp.weight q5_K +output.weight q8_0 +blk.1.ffn_down_exps.weight q5_1 +blk.1.ffn_down_shexp.weight q5_1 +blk.3.ffn_down_exps.weight q5_1 +blk.3.ffn_down_shexp.weight q5_1 [IQ3_S] iq3_s output.weight q8_0 +token_embd.weight iq4_nl +blk.0.ssm_in.weight iq4_nl +blk.1.ffn_down_exps.weight iq4_nl +blk.1.ffn_down_shexp.weight iq4_nl +blk.1.ffn_up_exps.weight iq4_nl +blk.1.ffn_up_shexp.weight iq4_nl +blk.2.ssm_in.weight iq4_nl +blk.3.ffn_down_exps.weight iq4_nl +blk.3.ffn_down_shexp.weight iq4_nl +blk.3.ffn_up_exps.weight iq4_nl +blk.3.ffn_up_shexp.weight iq4_nl +blk.4.ssm_in.weight iq4_nl +blk.5.attn_k.weight iq4_nl +blk.5.attn_q.weight iq4_nl +blk.5.attn_v.weight iq4_nl +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_down_shexp.weight iq4_nl +blk.6.ffn_up_exps.weight iq4_nl +blk.6.ffn_up_shexp.weight iq4_nl +blk.7.ssm_in.weight iq4_nl +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_down_shexp.weight iq4_nl +blk.8.ffn_up_exps.weight iq4_nl +blk.8.ffn_up_shexp.weight iq4_nl +blk.9.ssm_in.weight iq4_nl +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_down_shexp.weight iq4_nl +blk.10.ffn_up_exps.weight iq4_nl +blk.10.ffn_up_shexp.weight iq4_nl +blk.11.ssm_in.weight iq4_nl +blk.12.attn_k.weight iq4_nl +blk.12.attn_q.weight iq4_nl +blk.12.attn_v.weight iq4_nl +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_down_shexp.weight iq4_nl +blk.13.ffn_up_exps.weight iq4_nl +blk.13.ffn_up_shexp.weight iq4_nl +blk.14.ssm_in.weight iq4_nl +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_down_shexp.weight iq4_nl +blk.15.ffn_up_exps.weight iq4_nl +blk.15.ffn_up_shexp.weight iq4_nl +blk.16.ssm_in.weight iq4_nl +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_down_shexp.weight iq4_nl +blk.17.ffn_up_exps.weight iq4_nl +blk.17.ffn_up_shexp.weight iq4_nl +blk.18.ssm_in.weight iq4_nl +blk.19.attn_k.weight iq4_nl +blk.19.attn_q.weight iq4_nl +blk.19.attn_v.weight iq4_nl +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_down_shexp.weight iq4_nl +blk.20.ffn_up_exps.weight iq4_nl +blk.20.ffn_up_shexp.weight iq4_nl +blk.21.ssm_in.weight iq4_nl +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_down_shexp.weight iq4_nl +blk.22.ffn_up_exps.weight iq4_nl +blk.22.ffn_up_shexp.weight iq4_nl +blk.23.ssm_in.weight iq4_nl +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_down_shexp.weight iq4_nl +blk.24.ffn_up_exps.weight iq4_nl +blk.24.ffn_up_shexp.weight iq4_nl +blk.25.ssm_in.weight iq4_nl +blk.26.attn_k.weight iq4_nl +blk.26.attn_q.weight iq4_nl +blk.26.attn_v.weight iq4_nl +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_down_shexp.weight iq4_nl +blk.27.ffn_up_exps.weight iq4_nl +blk.27.ffn_up_shexp.weight iq4_nl +blk.28.ssm_in.weight iq4_nl +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_down_shexp.weight iq4_nl +blk.29.ffn_up_exps.weight iq4_nl +blk.29.ffn_up_shexp.weight iq4_nl +blk.30.ssm_in.weight iq4_nl +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_down_shexp.weight iq4_nl +blk.31.ffn_up_exps.weight iq4_nl +blk.31.ffn_up_shexp.weight iq4_nl +blk.32.ssm_in.weight iq4_nl +blk.33.attn_k.weight iq4_nl +blk.33.attn_q.weight iq4_nl +blk.33.attn_v.weight iq4_nl +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_down_shexp.weight iq4_nl +blk.34.ffn_up_exps.weight iq4_nl +blk.34.ffn_up_shexp.weight iq4_nl +blk.35.ssm_in.weight iq4_nl +blk.36.ffn_down_exps.weight iq4_nl +blk.36.ffn_down_shexp.weight iq4_nl +blk.36.ffn_up_exps.weight iq4_nl +blk.36.ffn_up_shexp.weight iq4_nl +blk.37.ssm_in.weight iq4_nl +blk.38.ffn_down_exps.weight iq4_nl +blk.38.ffn_down_shexp.weight iq4_nl +blk.38.ffn_up_exps.weight iq4_nl +blk.38.ffn_up_shexp.weight iq4_nl +blk.39.ssm_in.weight iq4_nl +blk.40.ffn_down_exps.weight iq4_nl +blk.40.ffn_down_shexp.weight iq4_nl +blk.40.ffn_up_exps.weight iq4_nl +blk.40.ffn_up_shexp.weight iq4_nl +blk.41.ssm_in.weight iq4_nl +blk.42.attn_k.weight iq4_nl +blk.42.attn_q.weight iq4_nl +blk.42.attn_v.weight iq4_nl +blk.43.ffn_down_exps.weight iq4_nl +blk.43.ffn_down_shexp.weight iq4_nl +blk.43.ffn_up_exps.weight iq4_nl +blk.43.ffn_up_shexp.weight iq4_nl +blk.44.ssm_in.weight iq4_nl +blk.45.ffn_down_exps.weight iq4_nl +blk.45.ffn_down_shexp.weight iq4_nl +blk.45.ffn_up_exps.weight iq4_nl +blk.45.ffn_up_shexp.weight iq4_nl +blk.46.ssm_in.weight iq4_nl +blk.47.ffn_down_exps.weight iq4_nl +blk.47.ffn_down_shexp.weight iq4_nl +blk.47.ffn_up_exps.weight iq4_nl +blk.47.ffn_up_shexp.weight iq4_nl +blk.48.ssm_in.weight iq4_nl +blk.49.ffn_down_exps.weight iq4_nl +blk.49.ffn_down_shexp.weight iq4_nl +blk.49.ffn_up_exps.weight iq4_nl +blk.49.ffn_up_shexp.weight iq4_nl +blk.50.ssm_in.weight iq4_nl +blk.51.ffn_down_exps.weight iq4_nl +blk.51.ffn_down_shexp.weight iq4_nl +blk.51.ffn_up_exps.weight iq4_nl +blk.51.ffn_up_shexp.weight iq4_nl [IQ3_M] iq3_s output.weight q8_0 -blk.1.ffn_down_exps.weight q4_K -blk.1.ffn_down_shexp.weight q4_K -blk.3.ffn_down_exps.weight q4_K -blk.3.ffn_down_shexp.weight q4_K +token_embd.weight iq4_nl +blk.0.ssm_in.weight iq4_nl +blk.1.ffn_down_exps.weight q5_0 +blk.1.ffn_down_shexp.weight q5_0 +blk.1.ffn_up_exps.weight iq4_nl +blk.1.ffn_up_shexp.weight iq4_nl +blk.2.ssm_in.weight iq4_nl +blk.3.ffn_down_exps.weight q5_0 +blk.3.ffn_down_shexp.weight q5_0 +blk.3.ffn_up_exps.weight iq4_nl +blk.3.ffn_up_shexp.weight iq4_nl +blk.4.ssm_in.weight iq4_nl +blk.5.attn_k.weight iq4_nl blk.5.attn_output.weight q4_K -blk.5.attn_v.weight q4_K +blk.5.attn_q.weight iq4_nl +blk.5.attn_v.weight q5_0 +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_down_shexp.weight iq4_nl +blk.6.ffn_up_exps.weight iq4_nl +blk.6.ffn_up_shexp.weight iq4_nl +blk.7.ssm_in.weight iq4_nl +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_down_shexp.weight iq4_nl +blk.8.ffn_up_exps.weight iq4_nl +blk.8.ffn_up_shexp.weight iq4_nl +blk.9.ssm_in.weight iq4_nl +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_down_shexp.weight iq4_nl +blk.10.ffn_up_exps.weight iq4_nl +blk.10.ffn_up_shexp.weight iq4_nl +blk.11.ssm_in.weight iq4_nl +blk.12.attn_k.weight iq4_nl blk.12.attn_output.weight q4_K -blk.12.attn_v.weight q4_K +blk.12.attn_q.weight iq4_nl +blk.12.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_down_shexp.weight iq4_nl +blk.13.ffn_up_exps.weight iq4_nl +blk.13.ffn_up_shexp.weight iq4_nl +blk.14.ssm_in.weight iq4_nl +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_down_shexp.weight iq4_nl +blk.15.ffn_up_exps.weight iq4_nl +blk.15.ffn_up_shexp.weight iq4_nl +blk.16.ssm_in.weight iq4_nl +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_down_shexp.weight iq4_nl +blk.17.ffn_up_exps.weight iq4_nl +blk.17.ffn_up_shexp.weight iq4_nl +blk.18.ssm_in.weight iq4_nl +blk.19.attn_k.weight iq4_nl blk.19.attn_output.weight q4_K -blk.19.attn_v.weight q4_K +blk.19.attn_q.weight iq4_nl +blk.19.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_down_shexp.weight iq4_nl +blk.20.ffn_up_exps.weight iq4_nl +blk.20.ffn_up_shexp.weight iq4_nl +blk.21.ssm_in.weight iq4_nl +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_down_shexp.weight iq4_nl +blk.22.ffn_up_exps.weight iq4_nl +blk.22.ffn_up_shexp.weight iq4_nl +blk.23.ssm_in.weight iq4_nl +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_down_shexp.weight iq4_nl +blk.24.ffn_up_exps.weight iq4_nl +blk.24.ffn_up_shexp.weight iq4_nl +blk.25.ssm_in.weight iq4_nl +blk.26.attn_k.weight iq4_nl blk.26.attn_output.weight q4_K -blk.26.attn_v.weight q4_K +blk.26.attn_q.weight iq4_nl +blk.26.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_down_shexp.weight iq4_nl +blk.27.ffn_up_exps.weight iq4_nl +blk.27.ffn_up_shexp.weight iq4_nl +blk.28.ssm_in.weight iq4_nl +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_down_shexp.weight iq4_nl +blk.29.ffn_up_exps.weight iq4_nl +blk.29.ffn_up_shexp.weight iq4_nl +blk.30.ssm_in.weight iq4_nl +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_down_shexp.weight iq4_nl +blk.31.ffn_up_exps.weight iq4_nl +blk.31.ffn_up_shexp.weight iq4_nl +blk.32.ssm_in.weight iq4_nl +blk.33.attn_k.weight iq4_nl blk.33.attn_output.weight q4_K -blk.33.attn_v.weight q4_K +blk.33.attn_q.weight iq4_nl +blk.33.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_down_shexp.weight iq4_nl +blk.34.ffn_up_exps.weight iq4_nl +blk.34.ffn_up_shexp.weight iq4_nl +blk.35.ssm_in.weight iq4_nl +blk.36.ffn_down_exps.weight iq4_nl +blk.36.ffn_down_shexp.weight iq4_nl +blk.36.ffn_up_exps.weight iq4_nl +blk.36.ffn_up_shexp.weight iq4_nl +blk.37.ssm_in.weight iq4_nl +blk.38.ffn_down_exps.weight iq4_nl +blk.38.ffn_down_shexp.weight iq4_nl +blk.38.ffn_up_exps.weight iq4_nl +blk.38.ffn_up_shexp.weight iq4_nl +blk.39.ssm_in.weight iq4_nl +blk.40.ffn_down_exps.weight iq4_nl +blk.40.ffn_down_shexp.weight iq4_nl +blk.40.ffn_up_exps.weight iq4_nl +blk.40.ffn_up_shexp.weight iq4_nl +blk.41.ssm_in.weight iq4_nl +blk.42.attn_k.weight iq4_nl blk.42.attn_output.weight q4_K -blk.42.attn_v.weight q4_K +blk.42.attn_q.weight iq4_nl +blk.42.attn_v.weight q5_0 +blk.43.ffn_down_exps.weight iq4_nl +blk.43.ffn_down_shexp.weight iq4_nl +blk.43.ffn_up_exps.weight iq4_nl +blk.43.ffn_up_shexp.weight iq4_nl +blk.44.ssm_in.weight iq4_nl +blk.45.ffn_down_exps.weight iq4_nl +blk.45.ffn_down_shexp.weight iq4_nl +blk.45.ffn_up_exps.weight iq4_nl +blk.45.ffn_up_shexp.weight iq4_nl +blk.46.ssm_in.weight iq4_nl +blk.47.ffn_down_exps.weight iq4_nl +blk.47.ffn_down_shexp.weight iq4_nl +blk.47.ffn_up_exps.weight iq4_nl +blk.47.ffn_up_shexp.weight iq4_nl +blk.48.ssm_in.weight iq4_nl +blk.49.ffn_down_exps.weight iq4_nl +blk.49.ffn_down_shexp.weight iq4_nl +blk.49.ffn_up_exps.weight iq4_nl +blk.49.ffn_up_shexp.weight iq4_nl +blk.50.ssm_in.weight iq4_nl +blk.51.ffn_down_exps.weight iq4_nl +blk.51.ffn_down_shexp.weight iq4_nl +blk.51.ffn_up_exps.weight iq4_nl +blk.51.ffn_up_shexp.weight iq4_nl [IQ2_S] iq2_xs output.weight q8_0 -token_embd.weight iq3_s -blk.1.ffn_down_exps.weight iq3_s -blk.1.ffn_down_shexp.weight iq3_s -blk.3.ffn_down_exps.weight iq3_s -blk.3.ffn_down_shexp.weight iq3_s +token_embd.weight iq4_nl +blk.0.ssm_in.weight iq4_nl +blk.1.ffn_down_exps.weight iq4_nl +blk.1.ffn_down_shexp.weight iq4_nl +blk.1.ffn_up_exps.weight iq4_nl +blk.1.ffn_up_shexp.weight iq4_nl +blk.2.ssm_in.weight iq4_nl +blk.3.ffn_down_exps.weight iq4_nl +blk.3.ffn_down_shexp.weight iq4_nl +blk.3.ffn_up_exps.weight iq4_nl +blk.3.ffn_up_shexp.weight iq4_nl +blk.4.ssm_in.weight iq4_nl +blk.5.attn_k.weight iq4_nl blk.5.attn_output.weight iq3_s -blk.5.attn_v.weight q4_K -blk.6.ffn_down_exps.weight iq3_s -blk.6.ffn_down_shexp.weight iq3_s +blk.5.attn_q.weight iq4_nl +blk.5.attn_v.weight q5_0 +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_down_shexp.weight iq4_nl +blk.6.ffn_up_exps.weight iq4_nl +blk.6.ffn_up_shexp.weight iq4_nl +blk.7.ssm_in.weight iq4_nl +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_down_shexp.weight iq4_nl +blk.8.ffn_up_exps.weight iq4_nl +blk.8.ffn_up_shexp.weight iq4_nl +blk.9.ssm_in.weight iq4_nl +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_down_shexp.weight iq4_nl +blk.10.ffn_up_exps.weight iq4_nl +blk.10.ffn_up_shexp.weight iq4_nl +blk.11.ssm_in.weight iq4_nl +blk.12.attn_k.weight iq4_nl blk.12.attn_output.weight iq3_s -blk.12.attn_v.weight q4_K +blk.12.attn_q.weight iq4_nl +blk.12.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_down_shexp.weight iq4_nl +blk.13.ffn_up_exps.weight iq4_nl +blk.13.ffn_up_shexp.weight iq4_nl +blk.14.ssm_in.weight iq4_nl +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_down_shexp.weight iq4_nl +blk.15.ffn_up_exps.weight iq4_nl +blk.15.ffn_up_shexp.weight iq4_nl +blk.16.ssm_in.weight iq4_nl +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_down_shexp.weight iq4_nl +blk.17.ffn_up_exps.weight iq4_nl +blk.17.ffn_up_shexp.weight iq4_nl +blk.18.ssm_in.weight iq4_nl +blk.19.attn_k.weight iq4_nl blk.19.attn_output.weight iq3_s -blk.19.attn_v.weight q4_K +blk.19.attn_q.weight iq4_nl +blk.19.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_down_shexp.weight iq4_nl +blk.20.ffn_up_exps.weight iq4_nl +blk.20.ffn_up_shexp.weight iq4_nl +blk.21.ssm_in.weight iq4_nl +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_down_shexp.weight iq4_nl +blk.22.ffn_up_exps.weight iq4_nl +blk.22.ffn_up_shexp.weight iq4_nl +blk.23.ssm_in.weight iq4_nl +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_down_shexp.weight iq4_nl +blk.24.ffn_up_exps.weight iq4_nl +blk.24.ffn_up_shexp.weight iq4_nl +blk.25.ssm_in.weight iq4_nl +blk.26.attn_k.weight iq4_nl blk.26.attn_output.weight iq3_s -blk.26.attn_v.weight q4_K +blk.26.attn_q.weight iq4_nl +blk.26.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_down_shexp.weight iq4_nl +blk.27.ffn_up_exps.weight iq4_nl +blk.27.ffn_up_shexp.weight iq4_nl +blk.28.ssm_in.weight iq4_nl +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_down_shexp.weight iq4_nl +blk.29.ffn_up_exps.weight iq4_nl +blk.29.ffn_up_shexp.weight iq4_nl +blk.30.ssm_in.weight iq4_nl +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_down_shexp.weight iq4_nl +blk.31.ffn_up_exps.weight iq4_nl +blk.31.ffn_up_shexp.weight iq4_nl +blk.32.ssm_in.weight iq4_nl +blk.33.attn_k.weight iq4_nl blk.33.attn_output.weight iq3_s -blk.33.attn_v.weight q4_K +blk.33.attn_q.weight iq4_nl +blk.33.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_down_shexp.weight iq4_nl +blk.34.ffn_up_exps.weight iq4_nl +blk.34.ffn_up_shexp.weight iq4_nl +blk.35.ssm_in.weight iq4_nl +blk.36.ffn_down_exps.weight iq4_nl +blk.36.ffn_down_shexp.weight iq4_nl +blk.36.ffn_up_exps.weight iq4_nl +blk.36.ffn_up_shexp.weight iq4_nl +blk.37.ssm_in.weight iq4_nl +blk.38.ffn_down_exps.weight iq4_nl +blk.38.ffn_down_shexp.weight iq4_nl +blk.38.ffn_up_exps.weight iq4_nl +blk.38.ffn_up_shexp.weight iq4_nl +blk.39.ssm_in.weight iq4_nl +blk.40.ffn_down_exps.weight iq4_nl +blk.40.ffn_down_shexp.weight iq4_nl +blk.40.ffn_up_exps.weight iq4_nl +blk.40.ffn_up_shexp.weight iq4_nl +blk.41.ssm_in.weight iq4_nl +blk.42.attn_k.weight iq4_nl blk.42.attn_output.weight iq3_s -blk.42.attn_v.weight q4_K +blk.42.attn_q.weight iq4_nl +blk.42.attn_v.weight q5_0 +blk.43.ffn_down_exps.weight iq4_nl +blk.43.ffn_down_shexp.weight iq4_nl +blk.43.ffn_up_exps.weight iq4_nl +blk.43.ffn_up_shexp.weight iq4_nl +blk.44.ssm_in.weight iq4_nl +blk.45.ffn_down_exps.weight iq4_nl +blk.45.ffn_down_shexp.weight iq4_nl +blk.45.ffn_up_exps.weight iq4_nl +blk.45.ffn_up_shexp.weight iq4_nl +blk.46.ssm_in.weight iq4_nl +blk.47.ffn_down_exps.weight iq4_nl +blk.47.ffn_down_shexp.weight iq4_nl +blk.47.ffn_up_exps.weight iq4_nl +blk.47.ffn_up_shexp.weight iq4_nl +blk.48.ssm_in.weight iq4_nl +blk.49.ffn_down_exps.weight iq4_nl +blk.49.ffn_down_shexp.weight iq4_nl +blk.49.ffn_up_exps.weight iq4_nl +blk.49.ffn_up_shexp.weight iq4_nl +blk.50.ssm_in.weight iq4_nl +blk.51.ffn_down_exps.weight iq4_nl +blk.51.ffn_down_shexp.weight iq4_nl +blk.51.ffn_up_exps.weight iq4_nl +blk.51.ffn_up_shexp.weight iq4_nl [IQ2_M] iq2_s output.weight q8_0 -token_embd.weight iq3_s -blk.1.ffn_down_exps.weight iq3_s -blk.1.ffn_down_shexp.weight iq3_s -blk.3.ffn_down_exps.weight iq3_s -blk.3.ffn_down_shexp.weight iq3_s +token_embd.weight iq4_nl +blk.0.ssm_in.weight iq4_nl +blk.1.ffn_down_exps.weight iq4_nl +blk.1.ffn_down_shexp.weight iq4_nl +blk.1.ffn_up_exps.weight iq4_nl +blk.1.ffn_up_shexp.weight iq4_nl +blk.2.ssm_in.weight iq4_nl +blk.3.ffn_down_exps.weight iq4_nl +blk.3.ffn_down_shexp.weight iq4_nl +blk.3.ffn_up_exps.weight iq4_nl +blk.3.ffn_up_shexp.weight iq4_nl +blk.4.ssm_in.weight iq4_nl +blk.5.attn_k.weight iq4_nl blk.5.attn_output.weight iq3_s -blk.5.attn_v.weight q4_K -blk.6.ffn_down_exps.weight iq3_s -blk.6.ffn_down_shexp.weight iq3_s +blk.5.attn_q.weight iq4_nl +blk.5.attn_v.weight q5_0 +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_down_shexp.weight iq4_nl +blk.6.ffn_up_exps.weight iq4_nl +blk.6.ffn_up_shexp.weight iq4_nl +blk.7.ssm_in.weight iq4_nl +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_down_shexp.weight iq4_nl +blk.8.ffn_up_exps.weight iq4_nl +blk.8.ffn_up_shexp.weight iq4_nl +blk.9.ssm_in.weight iq4_nl +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_down_shexp.weight iq4_nl +blk.10.ffn_up_exps.weight iq4_nl +blk.10.ffn_up_shexp.weight iq4_nl +blk.11.ssm_in.weight iq4_nl +blk.12.attn_k.weight iq4_nl blk.12.attn_output.weight iq3_s -blk.12.attn_v.weight q4_K +blk.12.attn_q.weight iq4_nl +blk.12.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_down_shexp.weight iq4_nl +blk.13.ffn_up_exps.weight iq4_nl +blk.13.ffn_up_shexp.weight iq4_nl +blk.14.ssm_in.weight iq4_nl +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_down_shexp.weight iq4_nl +blk.15.ffn_up_exps.weight iq4_nl +blk.15.ffn_up_shexp.weight iq4_nl +blk.16.ssm_in.weight iq4_nl +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_down_shexp.weight iq4_nl +blk.17.ffn_up_exps.weight iq4_nl +blk.17.ffn_up_shexp.weight iq4_nl +blk.18.ssm_in.weight iq4_nl +blk.19.attn_k.weight iq4_nl blk.19.attn_output.weight iq3_s -blk.19.attn_v.weight q4_K +blk.19.attn_q.weight iq4_nl +blk.19.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_down_shexp.weight iq4_nl +blk.20.ffn_up_exps.weight iq4_nl +blk.20.ffn_up_shexp.weight iq4_nl +blk.21.ssm_in.weight iq4_nl +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_down_shexp.weight iq4_nl +blk.22.ffn_up_exps.weight iq4_nl +blk.22.ffn_up_shexp.weight iq4_nl +blk.23.ssm_in.weight iq4_nl +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_down_shexp.weight iq4_nl +blk.24.ffn_up_exps.weight iq4_nl +blk.24.ffn_up_shexp.weight iq4_nl +blk.25.ssm_in.weight iq4_nl +blk.26.attn_k.weight iq4_nl blk.26.attn_output.weight iq3_s -blk.26.attn_v.weight q4_K +blk.26.attn_q.weight iq4_nl +blk.26.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_down_shexp.weight iq4_nl +blk.27.ffn_up_exps.weight iq4_nl +blk.27.ffn_up_shexp.weight iq4_nl +blk.28.ssm_in.weight iq4_nl +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_down_shexp.weight iq4_nl +blk.29.ffn_up_exps.weight iq4_nl +blk.29.ffn_up_shexp.weight iq4_nl +blk.30.ssm_in.weight iq4_nl +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_down_shexp.weight iq4_nl +blk.31.ffn_up_exps.weight iq4_nl +blk.31.ffn_up_shexp.weight iq4_nl +blk.32.ssm_in.weight iq4_nl +blk.33.attn_k.weight iq4_nl blk.33.attn_output.weight iq3_s -blk.33.attn_v.weight q4_K +blk.33.attn_q.weight iq4_nl +blk.33.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_down_shexp.weight iq4_nl +blk.34.ffn_up_exps.weight iq4_nl +blk.34.ffn_up_shexp.weight iq4_nl +blk.35.ssm_in.weight iq4_nl +blk.36.ffn_down_exps.weight iq4_nl +blk.36.ffn_down_shexp.weight iq4_nl +blk.36.ffn_up_exps.weight iq4_nl +blk.36.ffn_up_shexp.weight iq4_nl +blk.37.ssm_in.weight iq4_nl +blk.38.ffn_down_exps.weight iq4_nl +blk.38.ffn_down_shexp.weight iq4_nl +blk.38.ffn_up_exps.weight iq4_nl +blk.38.ffn_up_shexp.weight iq4_nl +blk.39.ssm_in.weight iq4_nl +blk.40.ffn_down_exps.weight iq4_nl +blk.40.ffn_down_shexp.weight iq4_nl +blk.40.ffn_up_exps.weight iq4_nl +blk.40.ffn_up_shexp.weight iq4_nl +blk.41.ssm_in.weight iq4_nl +blk.42.attn_k.weight iq4_nl blk.42.attn_output.weight iq3_s -blk.42.attn_v.weight q4_K +blk.42.attn_q.weight iq4_nl +blk.42.attn_v.weight q5_0 +blk.43.ffn_down_exps.weight iq4_nl +blk.43.ffn_down_shexp.weight iq4_nl +blk.43.ffn_up_exps.weight iq4_nl +blk.43.ffn_up_shexp.weight iq4_nl +blk.44.ssm_in.weight iq4_nl +blk.45.ffn_down_exps.weight iq4_nl +blk.45.ffn_down_shexp.weight iq4_nl +blk.45.ffn_up_exps.weight iq4_nl +blk.45.ffn_up_shexp.weight iq4_nl +blk.46.ssm_in.weight iq4_nl +blk.47.ffn_down_exps.weight iq4_nl +blk.47.ffn_down_shexp.weight iq4_nl +blk.47.ffn_up_exps.weight iq4_nl +blk.47.ffn_up_shexp.weight iq4_nl +blk.48.ssm_in.weight iq4_nl +blk.49.ffn_down_exps.weight iq4_nl +blk.49.ffn_down_shexp.weight iq4_nl +blk.49.ffn_up_exps.weight iq4_nl +blk.49.ffn_up_shexp.weight iq4_nl +blk.50.ssm_in.weight iq4_nl +blk.51.ffn_down_exps.weight iq4_nl +blk.51.ffn_down_shexp.weight iq4_nl +blk.51.ffn_up_exps.weight iq4_nl +blk.51.ffn_up_shexp.weight iq4_nl [IQ4_XS] iq4_xs output.weight q8_0 -blk.1.ffn_down_exps.weight q5_K -blk.1.ffn_down_shexp.weight q5_K -blk.3.ffn_down_exps.weight q5_K -blk.3.ffn_down_shexp.weight q5_K +token_embd.weight iq4_nl +blk.0.ssm_in.weight iq4_nl +blk.1.ffn_down_exps.weight q5_1 +blk.1.ffn_down_shexp.weight q5_1 +blk.1.ffn_up_exps.weight iq4_nl +blk.1.ffn_up_shexp.weight iq4_nl +blk.2.ssm_in.weight iq4_nl +blk.3.ffn_down_exps.weight q5_1 +blk.3.ffn_down_shexp.weight q5_1 +blk.3.ffn_up_exps.weight iq4_nl +blk.3.ffn_up_shexp.weight iq4_nl +blk.4.ssm_in.weight iq4_nl +blk.5.attn_k.weight iq4_nl +blk.5.attn_q.weight iq4_nl +blk.5.attn_v.weight iq4_nl +blk.6.ffn_down_exps.weight iq4_nl +blk.6.ffn_down_shexp.weight iq4_nl +blk.6.ffn_up_exps.weight iq4_nl +blk.6.ffn_up_shexp.weight iq4_nl +blk.7.ssm_in.weight iq4_nl +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_down_shexp.weight iq4_nl +blk.8.ffn_up_exps.weight iq4_nl +blk.8.ffn_up_shexp.weight iq4_nl +blk.9.ssm_in.weight iq4_nl +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_down_shexp.weight iq4_nl +blk.10.ffn_up_exps.weight iq4_nl +blk.10.ffn_up_shexp.weight iq4_nl +blk.11.ssm_in.weight iq4_nl +blk.12.attn_k.weight iq4_nl +blk.12.attn_q.weight iq4_nl +blk.12.attn_v.weight iq4_nl +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_down_shexp.weight iq4_nl +blk.13.ffn_up_exps.weight iq4_nl +blk.13.ffn_up_shexp.weight iq4_nl +blk.14.ssm_in.weight iq4_nl +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_down_shexp.weight iq4_nl +blk.15.ffn_up_exps.weight iq4_nl +blk.15.ffn_up_shexp.weight iq4_nl +blk.16.ssm_in.weight iq4_nl +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_down_shexp.weight iq4_nl +blk.17.ffn_up_exps.weight iq4_nl +blk.17.ffn_up_shexp.weight iq4_nl +blk.18.ssm_in.weight iq4_nl +blk.19.attn_k.weight iq4_nl +blk.19.attn_q.weight iq4_nl +blk.19.attn_v.weight iq4_nl +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_down_shexp.weight iq4_nl +blk.20.ffn_up_exps.weight iq4_nl +blk.20.ffn_up_shexp.weight iq4_nl +blk.21.ssm_in.weight iq4_nl +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_down_shexp.weight iq4_nl +blk.22.ffn_up_exps.weight iq4_nl +blk.22.ffn_up_shexp.weight iq4_nl +blk.23.ssm_in.weight iq4_nl +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_down_shexp.weight iq4_nl +blk.24.ffn_up_exps.weight iq4_nl +blk.24.ffn_up_shexp.weight iq4_nl +blk.25.ssm_in.weight iq4_nl +blk.26.attn_k.weight iq4_nl +blk.26.attn_q.weight iq4_nl +blk.26.attn_v.weight iq4_nl +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_down_shexp.weight iq4_nl +blk.27.ffn_up_exps.weight iq4_nl +blk.27.ffn_up_shexp.weight iq4_nl +blk.28.ssm_in.weight iq4_nl +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_down_shexp.weight iq4_nl +blk.29.ffn_up_exps.weight iq4_nl +blk.29.ffn_up_shexp.weight iq4_nl +blk.30.ssm_in.weight iq4_nl +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_down_shexp.weight iq4_nl +blk.31.ffn_up_exps.weight iq4_nl +blk.31.ffn_up_shexp.weight iq4_nl +blk.32.ssm_in.weight iq4_nl +blk.33.attn_k.weight iq4_nl +blk.33.attn_q.weight iq4_nl +blk.33.attn_v.weight iq4_nl +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_down_shexp.weight iq4_nl +blk.34.ffn_up_exps.weight iq4_nl +blk.34.ffn_up_shexp.weight iq4_nl +blk.35.ssm_in.weight iq4_nl +blk.36.ffn_down_exps.weight iq4_nl +blk.36.ffn_down_shexp.weight iq4_nl +blk.36.ffn_up_exps.weight iq4_nl +blk.36.ffn_up_shexp.weight iq4_nl +blk.37.ssm_in.weight iq4_nl +blk.38.ffn_down_exps.weight iq4_nl +blk.38.ffn_down_shexp.weight iq4_nl +blk.38.ffn_up_exps.weight iq4_nl +blk.38.ffn_up_shexp.weight iq4_nl +blk.39.ssm_in.weight iq4_nl +blk.40.ffn_down_exps.weight iq4_nl +blk.40.ffn_down_shexp.weight iq4_nl +blk.40.ffn_up_exps.weight iq4_nl +blk.40.ffn_up_shexp.weight iq4_nl +blk.41.ssm_in.weight iq4_nl +blk.42.attn_k.weight iq4_nl +blk.42.attn_q.weight iq4_nl +blk.42.attn_v.weight iq4_nl +blk.43.ffn_down_exps.weight iq4_nl +blk.43.ffn_down_shexp.weight iq4_nl +blk.43.ffn_up_exps.weight iq4_nl +blk.43.ffn_up_shexp.weight iq4_nl +blk.44.ssm_in.weight iq4_nl +blk.45.ffn_down_exps.weight iq4_nl +blk.45.ffn_down_shexp.weight iq4_nl +blk.45.ffn_up_exps.weight iq4_nl +blk.45.ffn_up_shexp.weight iq4_nl +blk.46.ssm_in.weight iq4_nl +blk.47.ffn_down_exps.weight iq4_nl +blk.47.ffn_down_shexp.weight iq4_nl +blk.47.ffn_up_exps.weight iq4_nl +blk.47.ffn_up_shexp.weight iq4_nl +blk.48.ssm_in.weight iq4_nl +blk.49.ffn_down_exps.weight iq4_nl +blk.49.ffn_down_shexp.weight iq4_nl +blk.49.ffn_up_exps.weight iq4_nl +blk.49.ffn_up_shexp.weight iq4_nl +blk.50.ssm_in.weight iq4_nl +blk.51.ffn_down_exps.weight iq4_nl +blk.51.ffn_down_shexp.weight iq4_nl +blk.51.ffn_up_exps.weight iq4_nl +blk.51.ffn_up_shexp.weight iq4_nl [IQ1_M] iq1_m output.weight q8_0 -token_embd.weight q2_K -blk.1.ffn_down_exps.weight q2_K -blk.1.ffn_down_shexp.weight q2_K -blk.3.ffn_down_exps.weight q2_K -blk.3.ffn_down_shexp.weight q2_K +token_embd.weight q4_0 +blk.0.ssm_in.weight iq4_nl +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_down_shexp.weight q4_0 +blk.1.ffn_up_exps.weight iq4_nl +blk.1.ffn_up_shexp.weight iq4_nl +blk.2.ssm_in.weight iq4_nl +blk.3.ffn_down_exps.weight q4_0 +blk.3.ffn_down_shexp.weight q4_0 +blk.3.ffn_up_exps.weight iq4_nl +blk.3.ffn_up_shexp.weight iq4_nl +blk.4.ssm_in.weight iq4_nl +blk.5.attn_k.weight iq4_nl blk.5.attn_output.weight iq2_xxs -blk.5.attn_v.weight q4_K -blk.6.ffn_down_exps.weight q2_K -blk.6.ffn_down_shexp.weight q2_K +blk.5.attn_q.weight iq4_nl +blk.5.attn_v.weight q5_0 +blk.6.ffn_down_exps.weight q4_0 +blk.6.ffn_down_shexp.weight q4_0 +blk.6.ffn_up_exps.weight iq4_nl +blk.6.ffn_up_shexp.weight iq4_nl +blk.7.ssm_in.weight iq4_nl +blk.8.ffn_down_exps.weight iq4_nl +blk.8.ffn_down_shexp.weight iq4_nl +blk.8.ffn_up_exps.weight iq4_nl +blk.8.ffn_up_shexp.weight iq4_nl +blk.9.ssm_in.weight iq4_nl +blk.10.ffn_down_exps.weight iq4_nl +blk.10.ffn_down_shexp.weight iq4_nl +blk.10.ffn_up_exps.weight iq4_nl +blk.10.ffn_up_shexp.weight iq4_nl +blk.11.ssm_in.weight iq4_nl +blk.12.attn_k.weight iq4_nl blk.12.attn_output.weight iq2_xxs -blk.12.attn_v.weight q4_K +blk.12.attn_q.weight iq4_nl +blk.12.attn_v.weight q5_0 +blk.13.ffn_down_exps.weight iq4_nl +blk.13.ffn_down_shexp.weight iq4_nl +blk.13.ffn_up_exps.weight iq4_nl +blk.13.ffn_up_shexp.weight iq4_nl +blk.14.ssm_in.weight iq4_nl +blk.15.ffn_down_exps.weight iq4_nl +blk.15.ffn_down_shexp.weight iq4_nl +blk.15.ffn_up_exps.weight iq4_nl +blk.15.ffn_up_shexp.weight iq4_nl +blk.16.ssm_in.weight iq4_nl +blk.17.ffn_down_exps.weight iq4_nl +blk.17.ffn_down_shexp.weight iq4_nl +blk.17.ffn_up_exps.weight iq4_nl +blk.17.ffn_up_shexp.weight iq4_nl +blk.18.ssm_in.weight iq4_nl +blk.19.attn_k.weight iq4_nl blk.19.attn_output.weight iq2_xxs -blk.19.attn_v.weight q4_K +blk.19.attn_q.weight iq4_nl +blk.19.attn_v.weight q5_0 +blk.20.ffn_down_exps.weight iq4_nl +blk.20.ffn_down_shexp.weight iq4_nl +blk.20.ffn_up_exps.weight iq4_nl +blk.20.ffn_up_shexp.weight iq4_nl +blk.21.ssm_in.weight iq4_nl +blk.22.ffn_down_exps.weight iq4_nl +blk.22.ffn_down_shexp.weight iq4_nl +blk.22.ffn_up_exps.weight iq4_nl +blk.22.ffn_up_shexp.weight iq4_nl +blk.23.ssm_in.weight iq4_nl +blk.24.ffn_down_exps.weight iq4_nl +blk.24.ffn_down_shexp.weight iq4_nl +blk.24.ffn_up_exps.weight iq4_nl +blk.24.ffn_up_shexp.weight iq4_nl +blk.25.ssm_in.weight iq4_nl +blk.26.attn_k.weight iq4_nl blk.26.attn_output.weight iq2_xxs -blk.26.attn_v.weight q4_K +blk.26.attn_q.weight iq4_nl +blk.26.attn_v.weight q5_0 +blk.27.ffn_down_exps.weight iq4_nl +blk.27.ffn_down_shexp.weight iq4_nl +blk.27.ffn_up_exps.weight iq4_nl +blk.27.ffn_up_shexp.weight iq4_nl +blk.28.ssm_in.weight iq4_nl +blk.29.ffn_down_exps.weight iq4_nl +blk.29.ffn_down_shexp.weight iq4_nl +blk.29.ffn_up_exps.weight iq4_nl +blk.29.ffn_up_shexp.weight iq4_nl +blk.30.ssm_in.weight iq4_nl +blk.31.ffn_down_exps.weight iq4_nl +blk.31.ffn_down_shexp.weight iq4_nl +blk.31.ffn_up_exps.weight iq4_nl +blk.31.ffn_up_shexp.weight iq4_nl +blk.32.ssm_in.weight iq4_nl +blk.33.attn_k.weight iq4_nl blk.33.attn_output.weight iq2_xxs -blk.33.attn_v.weight q4_K +blk.33.attn_q.weight iq4_nl +blk.33.attn_v.weight q5_0 +blk.34.ffn_down_exps.weight iq4_nl +blk.34.ffn_down_shexp.weight iq4_nl +blk.34.ffn_up_exps.weight iq4_nl +blk.34.ffn_up_shexp.weight iq4_nl +blk.35.ssm_in.weight iq4_nl +blk.36.ffn_down_exps.weight iq4_nl +blk.36.ffn_down_shexp.weight iq4_nl +blk.36.ffn_up_exps.weight iq4_nl +blk.36.ffn_up_shexp.weight iq4_nl +blk.37.ssm_in.weight iq4_nl +blk.38.ffn_down_exps.weight iq4_nl +blk.38.ffn_down_shexp.weight iq4_nl +blk.38.ffn_up_exps.weight iq4_nl +blk.38.ffn_up_shexp.weight iq4_nl +blk.39.ssm_in.weight iq4_nl +blk.40.ffn_down_exps.weight iq4_nl +blk.40.ffn_down_shexp.weight iq4_nl +blk.40.ffn_up_exps.weight iq4_nl +blk.40.ffn_up_shexp.weight iq4_nl +blk.41.ssm_in.weight iq4_nl +blk.42.attn_k.weight iq4_nl blk.42.attn_output.weight iq2_xxs -blk.42.attn_v.weight q4_K +blk.42.attn_q.weight iq4_nl +blk.42.attn_v.weight q5_0 +blk.43.ffn_down_exps.weight iq4_nl +blk.43.ffn_down_shexp.weight iq4_nl +blk.43.ffn_up_exps.weight iq4_nl +blk.43.ffn_up_shexp.weight iq4_nl +blk.44.ssm_in.weight iq4_nl +blk.45.ffn_down_exps.weight iq4_nl +blk.45.ffn_down_shexp.weight iq4_nl +blk.45.ffn_up_exps.weight iq4_nl +blk.45.ffn_up_shexp.weight iq4_nl +blk.46.ssm_in.weight iq4_nl +blk.47.ffn_down_exps.weight iq4_nl +blk.47.ffn_down_shexp.weight iq4_nl +blk.47.ffn_up_exps.weight iq4_nl +blk.47.ffn_up_shexp.weight iq4_nl +blk.48.ssm_in.weight iq4_nl +blk.49.ffn_down_exps.weight iq4_nl +blk.49.ffn_down_shexp.weight iq4_nl +blk.49.ffn_up_exps.weight iq4_nl +blk.49.ffn_up_shexp.weight iq4_nl +blk.50.ssm_in.weight iq4_nl +blk.51.ffn_down_exps.weight iq4_nl +blk.51.ffn_down_shexp.weight iq4_nl +blk.51.ffn_up_exps.weight iq4_nl +blk.51.ffn_up_shexp.weight iq4_nl [BF16] bf16 -output.weight q6_K [TQ1_0] tq1_0 output.weight q8_0 -token_embd.weight q4_K +token_embd.weight q5_0 +blk.0.ssm_in.weight q4_0 +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_down_shexp.weight q4_0 +blk.1.ffn_up_exps.weight q4_0 +blk.1.ffn_up_shexp.weight q4_0 +blk.2.ssm_in.weight q4_0 +blk.3.ffn_down_exps.weight q4_0 +blk.3.ffn_down_shexp.weight q4_0 +blk.3.ffn_up_exps.weight q4_0 +blk.3.ffn_up_shexp.weight q4_0 +blk.4.ssm_in.weight q4_0 +blk.5.attn_k.weight q4_0 +blk.5.attn_q.weight q4_0 +blk.5.attn_v.weight q4_0 +blk.6.ffn_down_exps.weight q4_0 +blk.6.ffn_down_shexp.weight q4_0 +blk.6.ffn_up_exps.weight q4_0 +blk.6.ffn_up_shexp.weight q4_0 +blk.7.ssm_in.weight q4_0 +blk.8.ffn_down_exps.weight q4_0 +blk.8.ffn_down_shexp.weight q4_0 +blk.8.ffn_up_exps.weight q4_0 +blk.8.ffn_up_shexp.weight q4_0 +blk.9.ssm_in.weight q4_0 +blk.10.ffn_down_exps.weight q4_0 +blk.10.ffn_down_shexp.weight q4_0 +blk.10.ffn_up_exps.weight q4_0 +blk.10.ffn_up_shexp.weight q4_0 +blk.11.ssm_in.weight q4_0 +blk.12.attn_k.weight q4_0 +blk.12.attn_q.weight q4_0 +blk.12.attn_v.weight q4_0 +blk.13.ffn_down_exps.weight q4_0 +blk.13.ffn_down_shexp.weight q4_0 +blk.13.ffn_up_exps.weight q4_0 +blk.13.ffn_up_shexp.weight q4_0 +blk.14.ssm_in.weight q4_0 +blk.15.ffn_down_exps.weight q4_0 +blk.15.ffn_down_shexp.weight q4_0 +blk.15.ffn_up_exps.weight q4_0 +blk.15.ffn_up_shexp.weight q4_0 +blk.16.ssm_in.weight q4_0 +blk.17.ffn_down_exps.weight q4_0 +blk.17.ffn_down_shexp.weight q4_0 +blk.17.ffn_up_exps.weight q4_0 +blk.17.ffn_up_shexp.weight q4_0 +blk.18.ssm_in.weight q4_0 +blk.19.attn_k.weight q4_0 +blk.19.attn_q.weight q4_0 +blk.19.attn_v.weight q4_0 +blk.20.ffn_down_exps.weight q4_0 +blk.20.ffn_down_shexp.weight q4_0 +blk.20.ffn_up_exps.weight q4_0 +blk.20.ffn_up_shexp.weight q4_0 +blk.21.ssm_in.weight q4_0 +blk.22.ffn_down_exps.weight q4_0 +blk.22.ffn_down_shexp.weight q4_0 +blk.22.ffn_up_exps.weight q4_0 +blk.22.ffn_up_shexp.weight q4_0 +blk.23.ssm_in.weight q4_0 +blk.24.ffn_down_exps.weight q4_0 +blk.24.ffn_down_shexp.weight q4_0 +blk.24.ffn_up_exps.weight q4_0 +blk.24.ffn_up_shexp.weight q4_0 +blk.25.ssm_in.weight q4_0 +blk.26.attn_k.weight q4_0 +blk.26.attn_q.weight q4_0 +blk.26.attn_v.weight q4_0 +blk.27.ffn_down_exps.weight q4_0 +blk.27.ffn_down_shexp.weight q4_0 +blk.27.ffn_up_exps.weight q4_0 +blk.27.ffn_up_shexp.weight q4_0 +blk.28.ssm_in.weight q4_0 +blk.29.ffn_down_exps.weight q4_0 +blk.29.ffn_down_shexp.weight q4_0 +blk.29.ffn_up_exps.weight q4_0 +blk.29.ffn_up_shexp.weight q4_0 +blk.30.ssm_in.weight q4_0 +blk.31.ffn_down_exps.weight q4_0 +blk.31.ffn_down_shexp.weight q4_0 +blk.31.ffn_up_exps.weight q4_0 +blk.31.ffn_up_shexp.weight q4_0 +blk.32.ssm_in.weight q4_0 +blk.33.attn_k.weight q4_0 +blk.33.attn_q.weight q4_0 +blk.33.attn_v.weight q4_0 +blk.34.ffn_down_exps.weight q4_0 +blk.34.ffn_down_shexp.weight q4_0 +blk.34.ffn_up_exps.weight q4_0 +blk.34.ffn_up_shexp.weight q4_0 +blk.35.ssm_in.weight q4_0 +blk.36.ffn_down_exps.weight q4_0 +blk.36.ffn_down_shexp.weight q4_0 +blk.36.ffn_up_exps.weight q4_0 +blk.36.ffn_up_shexp.weight q4_0 +blk.37.ssm_in.weight q4_0 +blk.38.ffn_down_exps.weight q4_0 +blk.38.ffn_down_shexp.weight q4_0 +blk.38.ffn_up_exps.weight q4_0 +blk.38.ffn_up_shexp.weight q4_0 +blk.39.ssm_in.weight q4_0 +blk.40.ffn_down_exps.weight q4_0 +blk.40.ffn_down_shexp.weight q4_0 +blk.40.ffn_up_exps.weight q4_0 +blk.40.ffn_up_shexp.weight q4_0 +blk.41.ssm_in.weight q4_0 +blk.42.attn_k.weight q4_0 +blk.42.attn_q.weight q4_0 +blk.42.attn_v.weight q4_0 +blk.43.ffn_down_exps.weight q4_0 +blk.43.ffn_down_shexp.weight q4_0 +blk.43.ffn_up_exps.weight q4_0 +blk.43.ffn_up_shexp.weight q4_0 +blk.44.ssm_in.weight q4_0 +blk.45.ffn_down_exps.weight q4_0 +blk.45.ffn_down_shexp.weight q4_0 +blk.45.ffn_up_exps.weight q4_0 +blk.45.ffn_up_shexp.weight q4_0 +blk.46.ssm_in.weight q4_0 +blk.47.ffn_down_exps.weight q4_0 +blk.47.ffn_down_shexp.weight q4_0 +blk.47.ffn_up_exps.weight q4_0 +blk.47.ffn_up_shexp.weight q4_0 +blk.48.ssm_in.weight q4_0 +blk.49.ffn_down_exps.weight q4_0 +blk.49.ffn_down_shexp.weight q4_0 +blk.49.ffn_up_exps.weight q4_0 +blk.49.ffn_up_shexp.weight q4_0 +blk.50.ssm_in.weight q4_0 +blk.51.ffn_down_exps.weight q4_0 +blk.51.ffn_down_shexp.weight q4_0 +blk.51.ffn_up_exps.weight q4_0 +blk.51.ffn_up_shexp.weight q4_0 [TQ2_0] tq2_0 output.weight q8_0 -token_embd.weight q4_K +token_embd.weight q5_0 +blk.0.ssm_in.weight q4_0 +blk.1.ffn_down_exps.weight q4_0 +blk.1.ffn_down_shexp.weight q4_0 +blk.1.ffn_up_exps.weight q4_0 +blk.1.ffn_up_shexp.weight q4_0 +blk.2.ssm_in.weight q4_0 +blk.3.ffn_down_exps.weight q4_0 +blk.3.ffn_down_shexp.weight q4_0 +blk.3.ffn_up_exps.weight q4_0 +blk.3.ffn_up_shexp.weight q4_0 +blk.4.ssm_in.weight q4_0 +blk.5.attn_k.weight q4_0 +blk.5.attn_q.weight q4_0 +blk.5.attn_v.weight q4_0 +blk.6.ffn_down_exps.weight q4_0 +blk.6.ffn_down_shexp.weight q4_0 +blk.6.ffn_up_exps.weight q4_0 +blk.6.ffn_up_shexp.weight q4_0 +blk.7.ssm_in.weight q4_0 +blk.8.ffn_down_exps.weight q4_0 +blk.8.ffn_down_shexp.weight q4_0 +blk.8.ffn_up_exps.weight q4_0 +blk.8.ffn_up_shexp.weight q4_0 +blk.9.ssm_in.weight q4_0 +blk.10.ffn_down_exps.weight q4_0 +blk.10.ffn_down_shexp.weight q4_0 +blk.10.ffn_up_exps.weight q4_0 +blk.10.ffn_up_shexp.weight q4_0 +blk.11.ssm_in.weight q4_0 +blk.12.attn_k.weight q4_0 +blk.12.attn_q.weight q4_0 +blk.12.attn_v.weight q4_0 +blk.13.ffn_down_exps.weight q4_0 +blk.13.ffn_down_shexp.weight q4_0 +blk.13.ffn_up_exps.weight q4_0 +blk.13.ffn_up_shexp.weight q4_0 +blk.14.ssm_in.weight q4_0 +blk.15.ffn_down_exps.weight q4_0 +blk.15.ffn_down_shexp.weight q4_0 +blk.15.ffn_up_exps.weight q4_0 +blk.15.ffn_up_shexp.weight q4_0 +blk.16.ssm_in.weight q4_0 +blk.17.ffn_down_exps.weight q4_0 +blk.17.ffn_down_shexp.weight q4_0 +blk.17.ffn_up_exps.weight q4_0 +blk.17.ffn_up_shexp.weight q4_0 +blk.18.ssm_in.weight q4_0 +blk.19.attn_k.weight q4_0 +blk.19.attn_q.weight q4_0 +blk.19.attn_v.weight q4_0 +blk.20.ffn_down_exps.weight q4_0 +blk.20.ffn_down_shexp.weight q4_0 +blk.20.ffn_up_exps.weight q4_0 +blk.20.ffn_up_shexp.weight q4_0 +blk.21.ssm_in.weight q4_0 +blk.22.ffn_down_exps.weight q4_0 +blk.22.ffn_down_shexp.weight q4_0 +blk.22.ffn_up_exps.weight q4_0 +blk.22.ffn_up_shexp.weight q4_0 +blk.23.ssm_in.weight q4_0 +blk.24.ffn_down_exps.weight q4_0 +blk.24.ffn_down_shexp.weight q4_0 +blk.24.ffn_up_exps.weight q4_0 +blk.24.ffn_up_shexp.weight q4_0 +blk.25.ssm_in.weight q4_0 +blk.26.attn_k.weight q4_0 +blk.26.attn_q.weight q4_0 +blk.26.attn_v.weight q4_0 +blk.27.ffn_down_exps.weight q4_0 +blk.27.ffn_down_shexp.weight q4_0 +blk.27.ffn_up_exps.weight q4_0 +blk.27.ffn_up_shexp.weight q4_0 +blk.28.ssm_in.weight q4_0 +blk.29.ffn_down_exps.weight q4_0 +blk.29.ffn_down_shexp.weight q4_0 +blk.29.ffn_up_exps.weight q4_0 +blk.29.ffn_up_shexp.weight q4_0 +blk.30.ssm_in.weight q4_0 +blk.31.ffn_down_exps.weight q4_0 +blk.31.ffn_down_shexp.weight q4_0 +blk.31.ffn_up_exps.weight q4_0 +blk.31.ffn_up_shexp.weight q4_0 +blk.32.ssm_in.weight q4_0 +blk.33.attn_k.weight q4_0 +blk.33.attn_q.weight q4_0 +blk.33.attn_v.weight q4_0 +blk.34.ffn_down_exps.weight q4_0 +blk.34.ffn_down_shexp.weight q4_0 +blk.34.ffn_up_exps.weight q4_0 +blk.34.ffn_up_shexp.weight q4_0 +blk.35.ssm_in.weight q4_0 +blk.36.ffn_down_exps.weight q4_0 +blk.36.ffn_down_shexp.weight q4_0 +blk.36.ffn_up_exps.weight q4_0 +blk.36.ffn_up_shexp.weight q4_0 +blk.37.ssm_in.weight q4_0 +blk.38.ffn_down_exps.weight q4_0 +blk.38.ffn_down_shexp.weight q4_0 +blk.38.ffn_up_exps.weight q4_0 +blk.38.ffn_up_shexp.weight q4_0 +blk.39.ssm_in.weight q4_0 +blk.40.ffn_down_exps.weight q4_0 +blk.40.ffn_down_shexp.weight q4_0 +blk.40.ffn_up_exps.weight q4_0 +blk.40.ffn_up_shexp.weight q4_0 +blk.41.ssm_in.weight q4_0 +blk.42.attn_k.weight q4_0 +blk.42.attn_q.weight q4_0 +blk.42.attn_v.weight q4_0 +blk.43.ffn_down_exps.weight q4_0 +blk.43.ffn_down_shexp.weight q4_0 +blk.43.ffn_up_exps.weight q4_0 +blk.43.ffn_up_shexp.weight q4_0 +blk.44.ssm_in.weight q4_0 +blk.45.ffn_down_exps.weight q4_0 +blk.45.ffn_down_shexp.weight q4_0 +blk.45.ffn_up_exps.weight q4_0 +blk.45.ffn_up_shexp.weight q4_0 +blk.46.ssm_in.weight q4_0 +blk.47.ffn_down_exps.weight q4_0 +blk.47.ffn_down_shexp.weight q4_0 +blk.47.ffn_up_exps.weight q4_0 +blk.47.ffn_up_shexp.weight q4_0 +blk.48.ssm_in.weight q4_0 +blk.49.ffn_down_exps.weight q4_0 +blk.49.ffn_down_shexp.weight q4_0 +blk.49.ffn_up_exps.weight q4_0 +blk.49.ffn_up_shexp.weight q4_0 +blk.50.ssm_in.weight q4_0 +blk.51.ffn_down_exps.weight q4_0 +blk.51.ffn_down_shexp.weight q4_0 +blk.51.ffn_up_exps.weight q4_0 +blk.51.ffn_up_shexp.weight q4_0 [MXFP4_MOE] mxfp4 output.weight q8_0 diff --git a/tests/snapshots/qwen3-0.6b.schema b/tests/snapshots/qwen3-0.6b.schema index 5ada58c60e..fd994f2c1e 100644 --- a/tests/snapshots/qwen3-0.6b.schema +++ b/tests/snapshots/qwen3-0.6b.schema @@ -2,10 +2,8 @@ # n_embd=1024, n_ff=3072, n_vocab=151936, n_layer=28, n_head=16, n_head_kv=8 [F32] f32 -output.weight q6_K [F16] f16 -output.weight q6_K [Q4_0] q4_0 output.weight q6_K @@ -1013,7 +1011,6 @@ blk.27.attn_output.weight iq2_xxs blk.27.attn_v.weight q2_K [BF16] bf16 -output.weight q6_K [TQ1_0] tq1_0 output.weight q6_K diff --git a/tests/snapshots/qwen3-14b.schema b/tests/snapshots/qwen3-14b.schema index 010849a57b..2fdd908bd0 100644 --- a/tests/snapshots/qwen3-14b.schema +++ b/tests/snapshots/qwen3-14b.schema @@ -2,10 +2,8 @@ # n_embd=5120, n_ff=17408, n_vocab=151936, n_layer=40, n_head=40, n_head_kv=8 [F32] f32 -output.weight q6_K [F16] f16 -output.weight q6_K [Q4_0] q4_0 output.weight q6_K @@ -1613,7 +1611,6 @@ blk.39.attn_output.weight iq2_xxs blk.39.attn_v.weight q4_K [BF16] bf16 -output.weight q6_K [TQ1_0] tq1_0 output.weight q6_K diff --git a/tests/snapshots/qwen3-coder-next.schema b/tests/snapshots/qwen3-coder-next.schema index 4f12721f16..9bf16fca4b 100644 --- a/tests/snapshots/qwen3-coder-next.schema +++ b/tests/snapshots/qwen3-coder-next.schema @@ -2,10 +2,8 @@ # n_embd=2048, n_ff=5120, n_vocab=151936, n_layer=48, n_head=16, n_head_kv=2, n_expert=512 [F32] f32 -output.weight q6_K [F16] f16 -output.weight q6_K [Q4_0] q4_0 output.weight q6_K @@ -1790,7 +1788,6 @@ blk.47.attn_v.weight q4_K output.weight q5_K [BF16] bf16 -output.weight q6_K [TQ1_0] tq1_0 token_embd.weight q4_K diff --git a/tests/snapshots/qwen3.5-27b.schema b/tests/snapshots/qwen3.5-27b.schema index b5e07f01fc..4080205336 100644 --- a/tests/snapshots/qwen3.5-27b.schema +++ b/tests/snapshots/qwen3.5-27b.schema @@ -2,10 +2,8 @@ # n_embd=5120, n_ff=17408, n_vocab=248320, n_layer=64, n_head=24, n_head_kv=4 [F32] f32 -output.weight q6_K [F16] f16 -output.weight q6_K [Q4_0] q4_0 output.weight q6_K @@ -1898,7 +1896,6 @@ blk.63.attn_output.weight iq2_xxs blk.63.attn_v.weight q4_K [BF16] bf16 -output.weight q6_K [TQ1_0] tq1_0 output.weight q6_K diff --git a/tests/snapshots/qwen3.5-397b-a17b.schema b/tests/snapshots/qwen3.5-397b-a17b.schema index e62948dac8..d5056b9835 100644 --- a/tests/snapshots/qwen3.5-397b-a17b.schema +++ b/tests/snapshots/qwen3.5-397b-a17b.schema @@ -2,10 +2,8 @@ # n_embd=4096, n_ff=0, n_vocab=248320, n_layer=60, n_head=32, n_head_kv=2, n_expert=512 [F32] f32 -output.weight q6_K [F16] f16 -output.weight q6_K [Q4_0] q4_0 output.weight q6_K @@ -2205,7 +2203,6 @@ blk.59.attn_output.weight iq2_xxs blk.59.attn_v.weight q4_K [BF16] bf16 -output.weight q6_K [TQ1_0] tq1_0 output.weight q6_K diff --git a/tests/snapshots/step-3.5-flash.schema b/tests/snapshots/step-3.5-flash.schema index c9d1be0082..36a13b3de2 100644 --- a/tests/snapshots/step-3.5-flash.schema +++ b/tests/snapshots/step-3.5-flash.schema @@ -2,10 +2,8 @@ # n_embd=4096, n_ff=11264, n_vocab=128896, n_layer=45, n_head=64, n_head_kv=8, n_expert=288 [F32] f32 -output.weight q6_K [F16] f16 -output.weight q6_K [Q4_0] q4_0 output.weight q6_K @@ -2078,7 +2076,6 @@ blk.44.attn_output.weight iq2_xxs blk.44.attn_v.weight q4_K [BF16] bf16 -output.weight q6_K [TQ1_0] tq1_0 output.weight q6_K diff --git a/tests/test-quant-type-selection.cpp b/tests/test-quant-type-selection.cpp index d5e06173f6..fd4b81ac97 100644 --- a/tests/test-quant-type-selection.cpp +++ b/tests/test-quant-type-selection.cpp @@ -268,21 +268,20 @@ static std::vector> compute_quant_types(llama_ quantize_state_impl qs(mdl, &qparams); - std::vector names; - names.reserve(tensors.size()); - for (const auto & mt : tensors) { - names.push_back(mt.tensor->name); + std::vector metadata(tensors.size()); + for (size_t i = 0; i < tensors.size(); ++i) { + metadata[i].name = tensors[i].tensor->name; } - init_quantize_state_counters(qs, names); + init_quantize_state_counters(qs, metadata); ggml_type default_type = llama_ftype_get_default_type(ftype); std::vector> result; result.reserve(tensors.size()); - for (const auto & mt : tensors) { - ggml_type got = llama_tensor_get_type(qs, default_type, mt.tensor, ftype); - result.push_back({ mt.tensor->name, got }); + for (size_t i = 0; i < tensors.size(); ++i) { + ggml_type got = llama_tensor_get_type(qs, &qparams, tensors[i].tensor, default_type, metadata[i]); + result.push_back({ metadata[i].name, got }); } return result; @@ -408,7 +407,7 @@ static bool run_test_section(llama_model & mdl, } if (got != expected) { - printf(" FAIL %-50s expected %s, got %s\n", name.c_str(), ggml_type_name(expected), ggml_type_name(got)); + printf(" FAIL %-50s %-10s expected %s, got %s\n", name.c_str(), llama_ftype_to_name(section.ftype), ggml_type_name(expected), ggml_type_name(got)); all_pass = false; } } @@ -432,8 +431,7 @@ static int run_remote_tests(const std::string & snapshot_dir, const char * argv0 std::string name = model_name_from_repo(spec.repo); printf("=== %s ===\n", name.c_str()); - fprintf(stderr, "Fetching model metadata for %s from %s...\n", name.c_str(), spec.repo); - auto result = gguf_fetch_model_meta(spec.repo, spec.quant); + auto result = gguf_fetch_model_meta(spec.repo, spec.quant, "", false); if (!result.has_value()) { printf(" SKIP (could not fetch model metadata)\n\n"); total_skip++; @@ -506,5 +504,8 @@ int main(int argc, char ** argv) { return run_generate(snapshot_dir); } + // suppress llama log warnings during test (e.g. tensor type fallback messages) + llama_log_set([](enum ggml_log_level, const char *, void *) {}, nullptr); + return run_remote_tests(snapshot_dir, argv[0]); }