Upload grpo_training.py with huggingface_hub

grpo_training.py

@@ -68,8 +68,11 @@ config = GRPOConfig(
 )
 
 # Define reward function for GRPO
-def preference_reward_func(inputs, prompts, completions, completion_ids_list):
+def preference_reward_func(**kwargs):
     """Simple reward function based on response length preference"""
+    # Extract completions from kwargs
+    completions = kwargs.get('completions', [])
+
     rewards = []
     for completion in completions:
         # Prefer shorter, more concise responses (addressing verbosity issue)