Upload folder using huggingface_hub

modeling_custom.py

@@ -149,6 +149,22 @@ class Qwen2ForCausalLMWithReward(Qwen2ForCausalLM):
         
         return lm_output, success_scores
 
+    def generate(self, *args, **kwargs):
+        """
+        Override generate to ensure return_score=False during generation.
+        
+        The generate() method expects the forward pass to return only the model output,
+        not a tuple. We automatically set return_score=False here.
+        
+        Usage:
+            >>> inputs = tokenizer("Hello", return_tensors="pt")
+            >>> outputs = model.generate(**inputs, max_new_tokens=50)
+            >>> # No need to pass return_score=False manually!
+        """
+        # Force return_score=False during generation
+        kwargs['return_score'] = False
+        return super().generate(*args, **kwargs)
+
 
 # For AutoModel registration
 AutoModelForCausalLM = Qwen2ForCausalLMWithReward