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

fixup tensor_requires_imatrix

This commit is contained in:
ddh0 2026-02-11 21:03:34 -06:00
parent 3211a847ef
commit 55dbee2bbe
1 changed files with 11 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
src/llama-quant.cpp
View File

@ -480,19 +480,18 @@ static size_t llama_tensor_quantize_impl(enum ggml_type new_type, const float *
}
static bool tensor_requires_imatrix(const llama_model_quantize_params * params, const ggml_tensor * t, const ggml_type dst_type) {
if (!params->imatrix) {
if (
dst_type == GGML_TYPE_IQ2_XXS || dst_type == GGML_TYPE_IQ2_XS ||
dst_type == GGML_TYPE_IQ2_S || dst_type == GGML_TYPE_IQ1_S || (
dst_type == GGML_TYPE_IQ1_M && strcmp(t->name, "token_embd.weight") &&
strcmp(t->name, "output.weight")
) || (
dst_type == GGML_TYPE_Q2_K && params->ftype == LLAMA_FTYPE_MOSTLY_Q2_K_S &&
strcmp(t->name, "token_embd.weight") != 0
)
) return true;
if (dst_type == GGML_TYPE_IQ2_XXS || dst_type == GGML_TYPE_IQ2_XS ||
dst_type == GGML_TYPE_IQ2_S || dst_type == GGML_TYPE_IQ1_S || (
dst_type == GGML_TYPE_IQ1_M && strcmp(t->name, "token_embd.weight") &&
strcmp(t->name, "output.weight")
) || (
dst_type == GGML_TYPE_Q2_K && params->ftype == LLAMA_FTYPE_MOSTLY_Q2_K_S &&
strcmp(t->name, "token_embd.weight") != 0
)) {
return true;
} else {
return false;
}
return false;
}
static void llama_model_quantize_impl(const std::string & fname_inp, const std::string & fname_out, const llama_model_quantize_params * params) {