match

example_trainer/run.py

@@ -100,6 +100,20 @@ def main():
     parser.add_argument("--save-path", type=str, default="trained_model_checkpoints")
     parser.add_argument("--checkpoint-interval", type=int, default=3)
     
+    # === GRPO/PPO Hyperparameters ===
+    parser.add_argument(
+        "--kl-coef", type=float, default=0.1,
+        help="KL divergence penalty coefficient. Higher = more conservative updates (default: 0.1)",
+    )
+    parser.add_argument(
+        "--clip-eps", type=float, default=0.2,
+        help="PPO clipping epsilon. Clips ratio to [1-eps, 1+eps] (default: 0.2)",
+    )
+    parser.add_argument(
+        "--no-reference-logprobs", action="store_true",
+        help="Disable use of inference logprobs as reference policy (not recommended)",
+    )
+    
     # === vLLM Server ===
     parser.add_argument("--vllm-port", type=int, default=9001, help="Port for vLLM server")
     parser.add_argument("--gpu-memory-utilization", type=float, default=0.5, help="vLLM GPU memory fraction")
@@ -233,6 +247,11 @@ def main():
         use_wandb=args.use_wandb,
         wandb_project=args.wandb_project,
         checkpoint_interval=args.checkpoint_interval,
+        # GRPO hyperparameters
+        kl_coef=args.kl_coef,
+        clip_eps=args.clip_eps,
+        use_reference_logprobs=not args.no_reference_logprobs,
+        benchmark=True,  # Always show timing info
     )
     
     try: