vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

quantize: fix F16/F32 downcast to q6_K

Browse source
This commit is contained in:
Johannes Gäßler 2024-03-10 19:50:11 +01:00
parent b838b53ad6
commit 13a39058d3
1 changed files with 1 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

2
llama.cpp
View file

@ -11675,7 +11675,7 @@ static ggml_type get_k_quant_type(quantize_state_internal & qs, ggml_type new_ty
ftype == LLAMA_FTYPE_MOSTLY_IQ1_S || ftype == LLAMA_FTYPE_MOSTLY_IQ2_S || ftype == LLAMA_FTYPE_MOSTLY_IQ2_M) { ftype == LLAMA_FTYPE_MOSTLY_IQ1_S || ftype == LLAMA_FTYPE_MOSTLY_IQ2_S || ftype == LLAMA_FTYPE_MOSTLY_IQ2_M) {
new_type = GGML_TYPE_Q5_K; new_type = GGML_TYPE_Q5_K;
} }
else if (new_type != GGML_TYPE_Q8_0) { else if (new_type != GGML_TYPE_Q8_0 && new_type != GGML_TYPE_F16 && new_type != GGML_TYPE_F32) {
new_type = GGML_TYPE_Q6_K; new_type = GGML_TYPE_Q6_K;
} }
} else if (name == "token_embd.weight") { } else if (name == "token_embd.weight") {