diff --git a/agentlightning/verl/trainer.py b/agentlightning/verl/trainer.py
index 413a0a1cf..2f9cee4de 100644
--- a/agentlightning/verl/trainer.py
+++ b/agentlightning/verl/trainer.py
@@ -417,10 +417,17 @@ def _train_step(self, batch_dict: dict) -> dict:
                     print(batch.batch.keys())
                     inputs = self.tokenizer.batch_decode(batch.batch["prompts"], skip_special_tokens=True)
                     outputs = self.tokenizer.batch_decode(batch.batch["responses"], skip_special_tokens=True)
+                    sample_gts = [
+                        item.non_tensor_batch.get("reward_model", {}).get(
+                            "ground_truth", None
+                        )
+                        for item in batch
+                    ]
                     scores = batch.batch["token_level_scores"].sum(-1).cpu().tolist()
                     self._dump_generations(
                         inputs=inputs,
                         outputs=outputs,
+                        gts=sample_gts,
                         scores=scores,
                         reward_extra_infos_dict=reward_extra_infos_dict,
                         dump_path=rollout_data_dir,