happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

more fixes

This commit is contained in:
younesbelkada 2025-07-03 15:26:33 +04:00
parent 03568c9358
commit 0c93ef6a9c
2 changed files with 4 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/llama-memory-hybrid.cpp
View File

@ -32,7 +32,7 @@ llama_memory_hybrid::llama_memory_hybrid(
mem_attn(new llama_kv_cache_unified( mem_attn(new llama_kv_cache_unified(
model, model,
filter_attn == nullptr ? filter_attn == nullptr ?
[&](int32_t il) { return hparams.is_recurrent(il); } [&](int32_t il) { return !hparams.is_recurrent(il); }
: filter_attn, : filter_attn,
type_k, type_k,
type_v, type_v,

12
src/llama-model.cpp
View File

@ -15101,14 +15101,6 @@ llama_memory_i * llama_model::create_memory(const llama_memory_params & params,
cparams.n_ctx = GGML_PAD(cparams.n_ctx, padding); cparams.n_ctx = GGML_PAD(cparams.n_ctx, padding);
// -> attn_filter
// if falcon-h1 -> [&](int32_t il) { return true; }
// case LLM_ARCH_FALCON_H1:
// llama_memory_hybrid::layer_filter_cb filter_attn = [](int32_t /*il*/) { return true; };
// llama_memory_hybrid::layer_filter_cb filter_recr = [](int32_t /*il*/) { return true; };
// default:
// llama_memory_hybrid::layer_filter_cb filter_attn = nullptr;
// llama_memory_hybrid::layer_filter_cb filter_recr = nullptr;
res = new llama_memory_hybrid( res = new llama_memory_hybrid(
/* model */ *this, /* model */ *this,
@ -15123,7 +15115,9 @@ llama_memory_i * llama_model::create_memory(const llama_memory_params & params,
/* recurrent_type_v */ GGML_TYPE_F32, /* recurrent_type_v */ GGML_TYPE_F32,
/* recurrent_kv_size */ std::max((uint32_t) 1, cparams.n_seq_max), /* recurrent_kv_size */ std::max((uint32_t) 1, cparams.n_seq_max),
/* n_seq_max */ cparams.n_seq_max, /* n_seq_max */ cparams.n_seq_max,
/* offload */ cparams.offload_kqv); /* offload */ cparams.offload_kqv,
/* filter_attn */ (arch == LLM_ARCH_FALCON_H1) ? [&](int32_t) { return true; } : (llama_memory_hybrid::layer_filter_cb)nullptr,
/* filter_recr */ (arch == LLM_ARCH_FALCON_H1) ? [&](int32_t) { return true; } : (llama_memory_hybrid::layer_filter_cb)nullptr);
} else { } else {
const auto padding = llama_kv_cache_unified::get_padding(cparams); const auto padding = llama_kv_cache_unified::get_padding(cparams);