[QUESTION] Function calling finetuning dataset sample

[DongqiShen]

I want to fine-tune Qwen2 to support custom function calling, but I am stuck on how to format function calling.

I checked the implementation in qwen-agent, and the template is complicated.

Before the prompt is sent to LLM, the system prompt is as follows:

{
  "role": "system",
  "content": "You are a helpful assistant.

# Tools

## You have access to the following tools:

### get_current_weather

get_current_weather: Get the current weather
Parameters: {\"type\": \"object\", \"properties\": {\"location\": {\"type\": \"string\", \"description\": \"The city and state, e.g. San Francisco, CA\"}, \"format\": {\"type\": \"string\", \"enum\": [\"celsius\", \"fahrenheit\"], \"description\": \"The temperature unit to use. Infer this from the user's location.\"}}, \"required\": [\"location\", \"format\"]}

### get_current_time

get_current_time: Get the current time
Parameters: {\"type\": \"object\", \"properties\": {\"location\": {\"type\": \"string\", \"description\": \"The city and state, e.g. San Francisco, CA\"}}, \"required\": [\"location\"]}

## When you need to call a tool, please insert the following command in your reply, which can be called zero or multiple times according to your needs:

✿FUNCTION✿: The tool to use, should be one of [get_current_weather,get_current_time]
✿ARGS✿: The input of the tool
✿RESULT✿: Tool results
✿RETURN✿: Reply based on tool results. Images need to be rendered as ![](url)"
}

What does the Final prompt sent to LLM look like for my custom function calling dataset?

Assuming the above get_current_weather and get_current_time are the functions I want to fine-tune.

Can someone give me a sample? Big Thanks for that。

I checked the official script for function finetuning of qwen1.5, which used ReAct Prompting technology and is somewhat different from the current version, so I would like to understand the function calling finetune of qwen2.

Thanks.