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

Process bpw targets up to B/F16

This commit is contained in:
Ed Addario 2025-11-30 00:29:35 +00:00
parent b97cda6289
commit 37cf51ebd0
No known key found for this signature in database
GPG Key ID: E7875815A3230993
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/llama-quant.cpp
View File

@ -2089,7 +2089,7 @@ static void llama_model_quantize_impl(const std::string & fname_inp, const std::
new_type = default_type;
// get more optimal quantization type based on the tensor shape, layer, etc.
if (!params->pure && ggml_is_quantized(default_type)) {
if (!params->pure && (ggml_is_quantized(default_type) || params->target_bpw != -1.0f)) {
int fallback = qs.n_fallback;
new_type = llama_tensor_get_type(qs, new_type, tensor, ftype);