Giters

MikeWangWZHL / Solo-Performance-Prompting

Repo for paper "Unleashing Cognitive Synergy in Large Language Models: A Task-Solving Agent through Multi-Persona Self-Collaboration"

Home Page:https://arxiv.org/abs/2307.05300

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Use without Azure?

marco2meyer opened this issue · comments

commented

Thank you so much for open-sourcing this repo, this is great!

I am trying to use the repo without Azure, but with no success yet. You note in the Readme that you ran the model on Azure. Have you tested whether the code runs if you set the Azure flag to false? I am getting somewhat inscrutable error messages from the open-ai API:

INFO:openai:error_code=None error_message='Invalid URL (POST /v1/engines/gpt4-32k/chat/completions)' error_param=None error_type=invalid_request_error message='OpenAI API error received' stream_error=False

commented

Hi,

Thanks for the message. I guess one possible reason for causing this problem is that you might need to modify the engine name to fit your API deployment. e.g., modifying the MODEL variable in the .sh scripts, such as: https://github.com/MikeWangWZHL/Solo-Performance-Prompting/blob/e77d109f1c9a8498995a5592dc0a5ea744b4b0f8/scripts/trivia_creative_writing.sh#L2C14-L2C14

Please first make sure that you setup the configuration in: https://github.com/MikeWangWZHL/Solo-Performance-Prompting/blob/main/config_template.sh

And then use a valid engine/model name compatible with the API, for example, if you are using the OpenAI's official API then the model names here https://platform.openai.com/docs/models/gpt-4 would be some valid choices? (the one "gpt4-32k" ("devgpt4-32k") we are using is only for the Azure deployment that we are using)

So a possible quick fix might be: change the MODEL="gpt4-32k" to MODEL="gpt-4-32k"?

commented