diff --git a/egs/speech_llm/ASR_LLM/whisper_llm_zh/decode.py b/egs/speech_llm/ASR_LLM/whisper_llm_zh/decode.py
index 79b6a6097..6efee5e4d 100755
--- a/egs/speech_llm/ASR_LLM/whisper_llm_zh/decode.py
+++ b/egs/speech_llm/ASR_LLM/whisper_llm_zh/decode.py
@@ -296,6 +296,10 @@ def decode_one_batch(
     # hyps = to_simple(hyps)
     # hyps = [params.normalizer.normalize(hyp) for hyp in hyps]
     print(hyps)
+    texts = batch["supervisions"]["text"]
+    for i, text in enumerate(texts):
+        print(f"ref: {text}")
+        print(f"hyp: {hyps[i]}")
     return {"beam-search": hyps}
 
 
@@ -476,7 +480,8 @@ def main():
     
     if params.use_flash_attn:
         attn_implementation = "flash_attention_2"
-        torch_dtype=torch.bfloat16
+        # torch_dtype=torch.bfloat16
+        torch_dtype=torch.float16
 
     else:
         attn_implementation = "eager"