fix scoring

Author: A9isha

src/maxtext/trainers/post_train/rl/math_verify_pool.py

@@ -223,7 +223,8 @@ def math_verify_pool(tmvp_config, items, scores, timeout=15, num_procs=None, log
         try:
           # .get(0) returns immediately since ready() was true
           _, score = job.get(0)
-          scores[idx] = max(scores[idx], tmvp_config.reward_exact_answer)
+          if score > 0.0:
+            scores[idx] = max(scores[idx], tmvp_config.reward_exact_answer)
           cnt += 1
         except Exception as e:
           if log_fn is not None: