qwen2的function calling数据格式不对

[missley]

Reminder

• I have read the README and searched the existing issues.

System Info

• llamafactory version: 0.8.3.dev0
• Platform: Linux-4.19.36-vhulk1907.1.0.h1524.eulerosv2r8.aarch64-aarch64-with-glibc2.28
• Python version: 3.10.14
• PyTorch version: 2.1.0 (NPU)
• Transformers version: 4.41.2
• Datasets version: 2.18.0
• Accelerate version: 0.30.1
• PEFT version: 0.11.1
• TRL version: 0.9.4
• NPU type: Ascend910ProB
• CANN version: 8.0.RC1

Reproduction

curl --location --request POST 'http://xxx/qwen2/v1/chat/completions' \
--header 'Content-Type: application/json' \
--data-raw '{
    "messages": [
        {
            "content": "我的工号是xxxx",
            "role": "system"
        },
        {
            "content": "我的交付组长的考勤是否有迟到？",
            "role": "user"
        }
    ],
    "tools": [
        {
            "type": "function",
            "function": {
                "name": "获取员工的交付组长信息",
                "description": "获取员工的交付组长信息",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "工号": {
                            "type": "string",
                            "description": "员工工号"
                        }
                    },
                    "required": [
                        "工号"
                    ]
                }
            }
        },
        {
            "type": "function",
            "function": {
                "name": "查询考勤信息",
                "description": "查询考勤信息",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "工号": {
                            "type": "string",
                            "description": "员工工号"
                        }
                    },
                    "required": [
                        "工号"
                    ]
                }
            }
        }
    ],
    "model": "qwen2",
    "stream": false,
    "max_tokens": 2024,
    "temperature": 0.2,
    "top_p": 0.3
}'

response：

{
    "id": "chatcmpl-04688b25903e4bff9356ffd4da2b3343",
    "object": "chat.completion",
    "created": 1720766740,
    "model": "qwen2",
    "choices": [
        {
            "index": 0,
            "message": {
                "role": "assistant",
                "content": "Action: 查询考勤信息.\nAction Input: {\"工号\": \"xxxxxx\"}",
                "tool_calls": null
            },
            "finish_reason": "stop"
        }
    ],
    "usage": {
        "prompt_tokens": 189,
        "completion_tokens": 23,
        "total_tokens": 212
    }
}

Expected behavior

正常格式应该如glm4：

{
    "id": "chatcmpl-b52a8a637df84b7cb43285af6f241b1c",
    "object": "chat.completion",
    "created": 1720766757,
    "model": "EMPTY",
    "choices": [
        {
            "index": 0,
            "message": {
                "role": "assistant",
                "content": null,
                "tool_calls": [
                    {
                        "id": "call_b80cde8bbf8545cf903e01136ce5d9b0",
                        "type": "function",
                        "function": {
                            "name": "获取员工的交付组长信息",
                            "arguments": "{\"工号\": \"xxxxx\"}"
                        }
                    }
                ]
            },
            "finish_reason": "tool_calls"
        }
    ],
    "usage": {
        "prompt_tokens": 275,
        "completion_tokens": 16,
        "total_tokens": 291
    }
}

Others

No response

[xiaojun777-huang]

tool_utils.py文件中，Action： tool name (one of [{tool_names}]). 这段提示会有概率返回extract_tools无法解析的文本。你给定的文本中就出现了一个小数点.正则匹配出错了。建议将提示词修改为Action： one of [{tool_names}]。不过这段tools的逻辑还是问题蛮多的，我提交了一个issue，你可以参考一下。

[lwj2001]

tool_utils.py文件中，Action： tool name (one of [{tool_names}]). 这段提示会有概率返回extract_tools无法解析的文本。你给定的文本中就出现了一个小数点.正则匹配出错了。建议将提示词修改为Action： one of [{tool_names}]。不过这段tools的逻辑还是问题蛮多的，我提交了一个issue，你可以参考一下。

您好，这个 tool_utils.py 文件在哪呢，我想根据我的训练数据来修改正则匹配

[xiaojun777-huang]

llama factory源码

[lwj2001]

llama factory源码

了解！原来是我的版本太低了，hh

[hiyouga]

fixed

[xiaojun777-huang]

建议将tool name也去掉，我做的测试用例，Qwen2-7b-Instruct还是有挺高概率返回Action: tool name (plus_func) Action Input: {A:1,B:2}，这样的字符串。这个字符串extract_tools里的正则表达式无法正确处理。

[Imbernoulli]

请问在这个问题中哪里用到了这个repo？