Man2Dev
diff --git a/‎convert-hf-to-gguf.py
Copy file name to clipboardExpand all lines: convert-hf-to-gguf.py
+6Lines changed: 6 additions & 0 deletions b/‎convert-hf-to-gguf.py
Copy file name to clipboardExpand all lines: convert-hf-to-gguf.py
+6Lines changed: 6 additions & 0 deletions
@@ -2458,6 +2458,12 @@ def write_tensors(self):
         tensor_map = gguf.get_tensor_name_map(self.model_arch, block_count)
 
         for name, data_torch in self.get_tensors():
+            # lm_head is not used in llama.cpp, while autoawq will include this tensor in model
+            # To prevent errors, skip loading lm_head.weight.
+            if name == "lm_head.weight":
+                print(f"Skipping get tensor {name!r} in safetensors so that convert can end normally.")
+                continue
+
             old_dtype = data_torch.dtype
 
             # convert any unsupported data types to float32