0724_4

examples/arealite/clevr_count_70k_grpo.py

@@ -44,10 +44,8 @@ def clevr_count_70k_reward_fn(
 
     if sol.strip() == ans.strip():
         print(f"completions: {completions}, answer: {answer}")
-        if is_thinking:
-            return 1
-        else:
-            return 0.8
+        return 1
+
 
     if re.match(r"^\[\d+(\.\d+)?\]$", sol.strip()):
         return 0.05
@@ -185,6 +183,8 @@ def main(args):
             stats_tracker.scope("grpo_actor"),
         ):
             stats = actor.ppo_update(batch)
+            wandb.log({"actor_reward": stats[0]["grpo_actor/final_reward/avg"]})
+            wandb.log({"task_reward": stats[0]["grpo_actor/task_reward/avg"]})
             actor.step_lr_scheduler()
             log_gpu_stats("ppo update")
 

pyproject.toml

@@ -31,7 +31,7 @@ dependencies = [
     "huggingface_hub",
     "datasets",
     "accelerate",
-    "transformers>=4.53.1",
+    "transformers==4.53.1",
     
     # Scientific computing
     "numpy<2.0.0",