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

llama-quant : correct `n_attention_wv` usage (#20357) 

* llama-quant : correct `n_attention_wv` usage

In #19770, I introduced a regression in the way the
`quantize_state_impl` counter values were initialized. I was
incrementing and using `n_attention_wv` in the same loop, when it should
have been fixed by the time we're deciding tensor types in
`llama_tensor_get_type_impl` (for `use_more_bits`).

I never observed a difference in any of [my
tests](https://github.com/ggml-org/llama.cpp/pull/19770#issuecomment-4000424712)
- it was only after @bartowski kindly pointed this out that I realized
it was incorrect. (Thanks!)

* simplify
This commit is contained in:
ddh0 2026-03-10 14:43:29 -05:00 committed by GitHub
parent 90b2731894
commit 10e5b148b0
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 16 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
src/llama-quant.cpp
View File

@ -870,9 +870,6 @@ static void llama_model_quantize_impl(const std::string & fname_inp, const std::
quantize_state_impl qs(model, params);
// these need to be set to n_layer by default
qs.n_ffn_down = qs.n_ffn_gate = qs.n_ffn_up = (int)model.hparams.n_layer;
if (params->only_copy) {
ftype = ml.ftype;
}
@ -979,6 +976,22 @@ static void llama_model_quantize_impl(const std::string & fname_inp, const std::
// compute tensor metadata once and cache it
std::vector<tensor_metadata> 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;
@ -991,16 +1004,6 @@ static void llama_model_quantize_impl(const std::string & fname_inp, const std::
const struct ggml_tensor * tensor = it->tensor;
const std::string name = ggml_get_name(tensor);
metadata[i].category = tensor_get_category(name);
if (category_is_attn_v(metadata[i].category)) {
++qs.n_attention_wv;
}
if (tensor_name_match_output_weight(name.c_str())) {
qs.has_tied_embeddings = false;
}
uint16_t i_split = params->keep_split ? it->idx : 0;
if (!ctx_outs[i_split]) {
ctx_outs[i_split].reset(gguf_init_empty());