Use mistral as planner

[lingchensanwen]

Hi team,

Thank you for the great work. I tried to replicate the part for using mistral as planner and I noticed that in tool_agent.py and api.py The part of mistral is always loading gpt3.5 instead. I'm wondering if you did this on purpose? I tried to replace both for mistral for example

if model_name in ['mistral']:
            self.max_token_length = 30000
            self.llm = ChatOpenAI(temperature=0,
                     max_tokens=4096,
                     openai_api_key="token-abc123", 
                     openai_api_base="http://localhost:8000/v1", 
                     model_name="mistralai/Mistral-7B-Instruct-v0.3")

However the following process steps fail - parsing is fine but then it shows the output is not in json as sometime mistral_two-stage_results is "".

Did you meet it before? How did you solve this issue in your mistral planning part? Looking forward to hearing from you.

[hsaest]

Hi Yating,

Thanks for your interest in our work.

We actually do this on purpose, please refer to this.

I cannot assess what happened since I don’t see the detailed responses from the planner. However, if I remember correctly, Mistral cannot always deliver a plan, so you might want to check Mistral’s outputs.

Best, Jian

[lingchensanwen]

thank you Jian! I'll check what's wrong on the mistral side.