Merge pull request vllm-project#7 from vllm-project/fix_reference

fix reference code

Author: youkaichao

support_materials/1-inference/generate.py

@@ -122,7 +122,9 @@ def main(
     print("load model")
     load_model(
         model,
-        os.path.join(ckpt_path, f"model{rank}-mp{world_size}.safetensors"))
+        os.path.join(ckpt_path, f"model{rank}-mp{world_size}.safetensors"),
+        # use strict=False to fix the lm head dtype mismatch
+        strict=False)
     print("I'm DeepSeek 👋")
 
     if interactive: