关于call functions的方式

[minpod]

发现dbgpt里面的call function是通过把工具信息组装在prompt里面传递给大模型，通过限制返回格式为json来获取调用的工具信息，但其实openai的接口就已经有tool的入参
response = client.chat.completions.create(
model=model,
messages=messages,
tools=tools,
tool_choice=tool_choice,
)

请问为何不直接用openai接口的入参进行传递，而要自己组装到prompt里面呢

[linjing-lab]

If dbgbt is the subset of large range model architect and application of openai, then json format is enough for dbgbt. Some usage of fully support keywords need to select multiple combinations, not include in dbgbt, because openai is a huge company for string to numeric, it need once support all usages in a time.