<think>标签训练后，模型预测结果无<think>标签

[qubingxin]

Reminder

• I have read the above rules and searched the existing issues.

System Info

• llamafactory version: 0.9.2.dev0
• Platform: Linux-5.4.241-1-tlinux4-0017.7-x86_64-with-glibc2.35
• Python version: 3.10.14
• PyTorch version: 2.4.0+cu124 (GPU)
• Transformers version: 4.45.2
• Datasets version: 3.1.0
• Accelerate version: 1.2.0
• PEFT version: 0.12.0
• TRL version: 0.9.6
• GPU type: NVIDIA H20
• GPU number: 8
• GPU memory: 95.00GB
• DeepSpeed version: 0.15.4
• vLLM version: dev

训练数据如alpaca格式，在output中的思维过程前后增加和, 无system prompt

训练过程中loss正常，但用vllm部署服务后，用openai接口请求输出结果没有和，这里很奇怪为什么和不遵循？

在源码中看到这里对和</think>进行了特殊处理，请问是这个原因吗？把前后这2个标签丢掉了？
@DataClass
class FunctionFormatter(StringFormatter):
def post_init(self):
super().post_init()
self.tool_utils = get_tool_utils(self.tool_format)

@override
def apply(self, **kwargs) -> SLOTS:
    content: str = kwargs.pop("content")
    regex = re.compile(r"<think>(.*)</think>", re.DOTALL)
    thought = re.search(regex, content)
    if thought:
        content = content.replace(thought.group(0), "")

    functions: List["FunctionCall"] = []
    try:
        tool_calls = json.loads(content)
        if not isinstance(tool_calls, list):  # parallel function call
            tool_calls = [tool_calls]

        for tool_call in tool_calls:
            functions.append(
                FunctionCall(tool_call["name"], json.dumps(tool_call["arguments"], ensure_ascii=False))
            )

    except json.JSONDecodeError:
        raise RuntimeError(f"Invalid JSON format in function message: {str([content])}.")  # flat string

    function_str = self.tool_utils.function_formatter(functions)
    if thought:
        function_str = thought.group(1) + function_str

    return super().apply(content=function_str)

Reproduction

Put your message here.

Others

No response

[qubingxin]

追加，提问内容的“"和“”内容不见了

[qubingxin]

think标签不输出了，带尖括号