From 00e8c5c5f6253347c61e3b74965df3577cccfee6 Mon Sep 17 00:00:00 2001 From: Jan Ploski Date: Sun, 1 Oct 2023 00:49:13 +0200 Subject: [PATCH] mpt : quick fix to avoid "Strange model" warning when quantizing MPT models --- llama.cpp | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/llama.cpp b/llama.cpp index 108ae5eb4..a2bdd9d3d 100644 --- a/llama.cpp +++ b/llama.cpp @@ -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; } }