migrate from llamafactory's chattemplate to llm's chattemplate

codecritic/utils/vllm.py

@@ -70,7 +70,7 @@ def score_worker(cuda_device, prompts, model_path, score_token):
         for response in output.outputs:
             # response.logprobs: list[dict[int, Logprob]] https://github.com/vllm-project/vllm/blob/main/vllm/sequence.py
             sample_logprobs = response.logprobs
-            logprob = sample_logprobs[1].get(score_token)
+            logprob = sample_logprobs[0].get(score_token)
             newitem = item.copy()
             # model always return 4 tokens, ['\n', 'Yes'/'No', '\n', <EOT>]
             if logprob: