diff options
| author | Iwan Kawrakow <iwan.kawrakow@gmail.com> | 2025-02-20 17:42:07 +0200 |
|---|---|---|
| committer | Iwan Kawrakow <iwan.kawrakow@gmail.com> | 2025-02-20 17:42:07 +0200 |
| commit | 4b45b82e67d9362e7522e5c7107e9d99219e0432 (patch) | |
| tree | 726d7d4f6ec3ce53a14171d3db7d8a9531c22f75 /src/llama.cpp | |
| parent | a45da7bfbf75503fe9e5a2f675db7825afdc6310 (diff) | |
Honor attn_output specified in the command line also for low-bit quants
Diffstat (limited to 'src/llama.cpp')
| -rw-r--r-- | src/llama.cpp | 3 |
1 files changed, 2 insertions, 1 deletions
diff --git a/src/llama.cpp b/src/llama.cpp index 0257a0a3..28e887ee 100644 --- a/src/llama.cpp +++ b/src/llama.cpp @@ -16339,7 +16339,8 @@ static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type n ++qs.i_ffn_down; } else if (name.find("attn_output.weight") != std::string::npos) { - if (qs.model.hparams.n_expert >= 4) { + if (qs.params->attn_output_type < GGML_TYPE_COUNT) new_type = qs.params->attn_output_type; + else if (qs.model.hparams.n_expert >= 4) { new_type = GGML_TYPE_Q5_K; } else { if (ftype == LLAMA_FTYPE_MOSTLY_IQ1_S || ftype == LLAMA_FTYPE_MOSTLY_IQ1_M) new_type = GGML_TYPE_IQ2_K; |