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

Execute bpw_overrides() only if an imatrix file is provided

This commit is contained in:
Ed Addario 2025-08-24 13:36:03 +01:00
parent 3856d60328
commit 61c0e01f50
No known key found for this signature in database
GPG Key ID: E7875815A3230993
1 changed files with 12 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
src/llama-quant.cpp
View File

@ -1525,9 +1525,18 @@ static void llama_model_quantize_impl(const std::string & fname_inp, const std::
}
std::unordered_map<std::string, ggml_type> bpw_overrides = {};
if (params->target_bpw != -1.0f) {
LLAMA_LOG_INFO("%s: computing tensor quantization mix to achieve %.3f bpw at lowest ppl - this operation may take some time\n", __func__, params->target_bpw);
bpw_overrides = target_bpw_type(ml, read_data, model, tensors, mapped, values_data, activations_data, params, nthread);
if (params->target_bpw != -1.0f && !params->only_copy) {
if (params->imatrix) {
if (params->activations) {
LLAMA_LOG_INFO("%s: imatrix with activations provided, target bpw quantization will be more accurate\n", __func__);
} else {
LLAMA_LOG_WARN("%s: imatrix without activations provided, target bpw quantization will be less accurate\n", __func__);
}
LLAMA_LOG_INFO("%s: computing tensor quantization mix to achieve %.4f bpw\n", __func__, params->target_bpw);
bpw_overrides = target_bpw_type(ml, read_data, model, tensors, mapped, values_data, activations_data, params, nthread);
} else {
LLAMA_LOG_WARN("%s: no imatrix provided, target bpw will not apply\n", __func__);
}
}
int cur_split = -1;