LLM大语言模型（十三）：ChatGLM3-6B兼容Langchain的Function Call的一步一步的详细转换过程记录

# LangChain：原始prompt

System: Respond to the human as helpfully and accurately as possible. You have access to the following tools:

Calculator: Useful for when you need to calculate math problems, args: {\'calculation\': {\'description\': \'calculation to perform\', \'title\': \'Calculation\', \'type\': \'string\'}}

Use a json blob to specify a tool by providing an action key (tool name) and an action_input key (tool input).

Valid "action" values: "Final Answer" or Calculator

Provide only ONE action per $JSON_BLOB, as shown:

```
{
    "action": $TOOL_NAME,
    "action_input": $INPUT
}
```
Follow this format:

Question: input question to answer
Thought: consider previous and subsequent steps
Action:
```
$JSON_BLOB
```
Observation: action result
... (repeat Thought/Action/Observation N times)
Thought: I know what to respond
Action:
```
{
    "action": "Final Answer",
    "action_input": "Final response to human"
}

Begin! Reminder to ALWAYS respond with a valid json blob of a single action. Use tools if necessary. Respond directly if appropriate. Format is Action:```$JSON_BLOB```then Observation
Human: 34 * 34

(reminder to respond in a JSON blob no matter what)

# ChatGLM：找到原始prompt中关于tool的说明 

Calculator: Useful for when you need to calculate math problems, args: {'calculation': {'description': 'calculation to perform', 'title': 'Calculation', 'type': 'string'}}

# ChatGLM：找到原始prompt中用户输入

Human: 34 * 34\n\n\n(reminder to respond in a JSON blob no matter what)

# ChatGLM：将原始prompt转换为ChatGLM的会话格式，并记录到self.history，同时找到用户输入作为接下来的query=34 * 34

[
    {
        'role': 'system', 
        'content': 'Answer the following questions as best as you can. You have access to the following tools:', 
        'tools': [
            {
                'name': 'Calculator', 
                'description': 'Useful for when you need to calculate math problems', 
                'parameters': {
                    'calculation': {
                        'description': 'calculation to perform', 
                        'type': 'string'
                    }
                }
            }
        ]
    }, 
    {
        'role': 'user', 
        'content': '34 * 34\n\n\n (reminder to respond in a JSON blob no matter what)'
    }
]

# ChatGLM：依据self.history和query进行生成，生成结果赋值给self.history，新的self.history内容如下

[{'role': 'system', 'content': 'Answer the following questions as best as you can. You have access to the following tools:', 'tools': [{'name': 'Calculator', 'description': 'Useful for when you need to calculate math problems', 'parameters': {'calculation': {'description': 'calculation to perform', 'type': 'string'}}}]}, {'role': 'user', 'content': '34 * 34\n\n\n (reminder to respond in a JSON blob no matter what)'}, {'role': 'user', 'content': '34 * 34'}, {'role': 'assistant', 'metadata': 'Calculator', 'content': " ```python\ntool_call(calculation='34*34')\n```"}]

==新增了两条信息==

{'role': 'user', 'content': '34 * 34'}, 
{'role': 'assistant', 'metadata': 'Calculator', 'content': " ```python\ntool_call(calculation='34*34')\n```"}

# ChatGLM：解析LLM最新回答中的tool，并作为_call()函数的返回

response = '\nAction: \n```\n{"action": "Calculator", "action_input": {"calculation": "34*34"}}\n```'

# ChatGLM：更新_call()的入参History，增加一个pair=(prompt,response)，传递给LangChain

==此时prompt就是原始prompt==
==response就是ChatGLM生成的接下来要用到的Tool，也就是原始prompt里希望LLM返回的结果==

# LangChain：执行Tool的调用，得到Tool的返回值，继续调用LLM

==这时候LLM还没有返回Final answer，所以要继续执行LLM==

# ChatGLM：此时的prompt是在原始prompt基础上再增加了上一步Tool的调用信息

'System: Respond to the human as helpfully and accurately as possible. You have access to the following tools:\n\nCalculator: Useful for when you need to calculate math problems, args: {\'calculation\': {\'description\': \'calculation to perform\', \'title\': \'Calculation\', \'type\': \'string\'}}\n\nUse a json blob to specify a tool by providing an action key (tool name) and an action_input key (tool input).\n\nValid "action" values: "Final Answer" or Calculator\n\nProvide only ONE action per $JSON_BLOB, as shown:\n\n```\n{\n  "action": $TOOL_NAME,\n  "action_input": $INPUT\n}\n```\n\nFollow this format:\n\nQuestion: input question to answer\nThought: consider previous and subsequent steps\nAction:\n```\n$JSON_BLOB\n```\nObservation: action result\n... (repeat Thought/Action/Observation N times)\nThought: I know what to respond\nAction:\n```\n{\n  "action": "Final Answer",\n  "action_input": "Final response to human"\n}\n\nBegin! Reminder to ALWAYS respond with a valid json blob of a single action. Use tools if necessary. Respond directly if appropriate. Format is Action:```$JSON_BLOB```then Observation\nHuman: 34 * 34\n\n\n

Action: \n```\n{"action": "Calculator", "action_input": {"calculation": "34*34"}}\n```\nObservation: 1156\nThought: \n 
==这一段是新增的，增加了上一步Action的Tool的执行结果==

(reminder to respond in a JSON blob no matter what)'

# ChatGLM解析新prompt中的observation

得到1156
向self.history新增一条信息：
{'role': 'observation', 'content': '1156'}

# ChatGLM：再次执行chat，进行生成

入参：此时query是空，history是所有的历史
返回结果，新增如下两条信息：
{'role': 'user', 'content': ''}
{'role': 'assistant', 'metadata': '', 'content': '{\n    " calculation": "34*34",\n    " result": 1156\n}'}

# ChatGLM：解析tool，发现self.history里最后一条消息的metadata是空，说明没有tool需要调用了，可以拼接Final answer，_call()返回值如下

response = '\nAction: \n```\n{"action": "Final Answer", "action_input": "{\\n    \\" calculation\\": \\"34*34\\",\\n    \\" result\\": 1156\\n}"}\n```'

# ChatGLM：_call()向入参的History里增加了一个新的pair

0=新的prompt
1=response

# LangChain：收到了Final Answer，调用结束，最后输出

{'input': '34 * 34', 'output': '{\n    " calculation": "34*34",\n    " result": 1156\n}'}

 参考

1. LLM大语言模型（十二）：关于ChatGLM3-6B不兼容Langchain 的Function Call-CSDN博客
2.  LLM大语言模型（十一）：基于自定义的ChatGLM3-6B构建LangChain的chain-CSDN博客
3. LLM大语言模型（十）：LangChain自定义Agent使用自定义的LLM-CSDN博客
4. LLM大语言模型（九）：LangChain封装自定义的LLM-CSDN博客
5. LLM大语言模型（八）：ChatGLM3-6B使用的tokenizer模型BAAI/bge-large-zh-v1.5-CSDN博客
6. LLM大语言模型（七）：部署ChatGLM3-6B并提供HTTP server能力
7. LLM大语言模型（四）：在ChatGLM3-6B中使用langchain_chatglm3-6b langchain-CSDN博客