diff options
| author | Nexesenex <124105151+Nexesenex@users.noreply.github.com> | 2024-03-22 14:32:02 +0100 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2024-03-22 15:32:02 +0200 |
| commit | e80f06d2a194be62ab5b1cd7ef7c7a5b241dd4fb (patch) | |
| tree | ecd5d09b0c627826dd50d419f8cec057c4805d11 /llama.cpp | |
| parent | f77a8ffd3bbde77b7819823b0c006fd8c2d5cae4 (diff) | |
llama : correction of the attn.v.weight quantization for IQ3_XS (#6209)
IQ3_XS was not mentioned, IQ3_S and IQ3_M were present twice.
That PR corrects this in the manner which was probably intended initially.
Diffstat (limited to 'llama.cpp')
| -rw-r--r-- | llama.cpp | 8 |
1 files changed, 1 insertions, 7 deletions
@@ -12027,13 +12027,7 @@ static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type n else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_XXS) { new_type = qs.model.hparams.n_gqa() >= 4 ? GGML_TYPE_Q4_K : !qs.has_imatrix ? GGML_TYPE_IQ3_S : GGML_TYPE_IQ3_XXS; } - else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_S && qs.model.hparams.n_gqa() >= 4) { - new_type = GGML_TYPE_Q4_K; - } - else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_M) { - new_type = GGML_TYPE_Q4_K; - } - else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_S && qs.model.hparams.n_gqa() >= 4) { + else if ((ftype == LLAMA_FTYPE_MOSTLY_IQ3_XS || ftype == LLAMA_FTYPE_MOSTLY_IQ3_S) && qs.model.hparams.n_gqa() >= 4) { new_type = GGML_TYPE_Q4_K; } else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_M) { |