浅谈OpenAIClaude LLM Tools的额外配置
前言
- https://platform.openai.com/docs/guides/function-calling?api-mode=chat&strict-mode=disabled#additional-configurations
- https://docs.anthropic.com/en/docs/build-with-claude/tool-use/overview#forcing-tool-use
tool_choice
- “none” 就是不用tools， tools传入也为味空
- auto，就是可能调用tool，或者不调用，agent loop，很多写的范式都是，调用tool，就执行拿到结果进入下一个loop，如果不调用tool就直接
- required：必须使用其中一个到多个工具，这样的agentloop可能就是tools里有对应的类似于stopTool，调用到就agent ending
- forced function： 只调用一个特定tool。
claude也是一样的，就是传入参数 required 为any了
parallel_tool_calls
parallel_tool_calls = True的时候返回的ChatCompletionMessage中的choice.message.tool_calls有多个 toolcall
为False的时候， 只能每次调用一个tool
Strict mode
“strict”= true的时候更能保证json output的输出，这个参数感觉大家用的很少，实际上为true和false，对于tool的parameters写法很多都可以等价
比如：
{"type": "function","function": {"name": "get_weather","description": "Retrieves current weather for the given location.","parameters": {"type": "object","properties": {"location": {"type": "string","description": "City and country e.g. Bogotá, Colombia"},"units": {"type": "string","enum": ["celsius", "fahrenheit"],"description": "Units the temperature will be returned in."}},"required": ["location"],}}
}
这个是我们平时很爱写的写法，看到的基本上都是这种，只会把一定需要的参数写为required中，但openai的"strict"= true， 对type做一下修改是一样的效果
{"type": "function","function": {"name": "get_weather","description": "Retrieves current weather for the given location.","strict": true,"parameters": {"type": "object","properties": {"location": {"type": "string","description": "City and country e.g. Bogotá, Colombia"},"units": {"type": ["string", "null"],"enum": ["celsius", "fahrenheit"],"description": "Units the temperature will be returned in."}},"required": ["location", "units"],"additionalProperties": false}}
}
加入 "null"的type