diff --git a/llama.cpp b/llama.cpp
index 2f61faa9b..a76d5f7ec 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -13717,7 +13717,7 @@ 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 || name.find("attn_qkv.weight") != std::string::npos) {
+        if (name.find("attn_v.weight") != std::string::npos || name.find("attn_qkv.weight") != std::string::npos || name.find("attn.Wqkv.weight") != std::string::npos) {
             ++qs.n_attention_wv;
         } else if (name == LLM_TN(model.arch)(LLM_TENSOR_OUTPUT, "weight")) {
             qs.has_output = true;