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

mpt : quick fix to avoid "Strange model" warning when quantizing MPT models

Browse source
This commit is contained in:
Jan Ploski 2023-10-01 00:49:13 +02:00
parent 84e30e891c
commit 00e8c5c5f6
1 changed files with 4 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

6
llama.cpp
View file

@ -6511,10 +6511,12 @@ static void llama_model_quantize_internal(const std::string & fname_inp, const s
const std::string name = ggml_get_name(meta);
// TODO: avoid hardcoded tensor names - use the TN_* constants
if (name.find("attn_v.weight") != std::string::npos) {
if (name.find("attn_v.weight") != std::string::npos ||
name.find("attn.Wqkv.weight") != std::string::npos) {
++n_attention_wv;
}
else if (name.find("ffn_down.weight") != std::string::npos) {
else if (name.find("ffn_down.weight") != std::string::npos ||
name.find("ffn.down_proj.weight") != std::string::npos) {
++n_feed_forward_w2;
}
}