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

better sanity check skipping for QRWKV6 in llama-quant

Browse source 
thanks @compilade

Signed-off-by: Molly Sophia <mollysophia379@gmail.com>
This commit is contained in:
Molly Sophia 2025-01-10 09:42:46 +08:00
parent d8a304c2ef
commit 324afba5cc
1 changed files with 2 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

4
src/llama-quant.cpp
View file

@ -621,8 +621,8 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
qs.n_ffn_down = qs.n_ffn_gate = qs.n_ffn_up = (int)model.hparams.n_layer; qs.n_ffn_down = qs.n_ffn_gate = qs.n_ffn_up = (int)model.hparams.n_layer;
// sanity checks // sanity checks for models that have attention layers
if (!llama_model_is_recurrent(&model)) if (qs.n_attention_wv != 0)
{ {
const auto & n_head_kv_iter = model.hparams.n_head_kv_arr.begin(); const auto & n_head_kv_iter = model.hparams.n_head_kv_arr.begin();
// attention layers have a non-zero number of kv heads // attention layers have a non-zero number of kv heads