Incorrect response format of "tool_calls"
#13
by
shaunxu
- opened
The response format of "tool_calls" was different from Qwen-Max, Qwen-Plus, etc., also different from OpenAI anq other LLM such as Doubao, Moonshot. It respond tool calls as a regular content wrapped by "" and "". For example something like this
<tool_call>
{"name": "get_weather", "arguments": {"city": "Beijing"}}
</tool_call>