[Feature Request] Critic selection mechanism for models don't support multiple response at one time #996
Assignees
Labels
call for contribution
enhancement
New feature or request
New Feature
P1
Task with middle level priority
Milestone
Comments
Wendong-Fan
added
enhancement
|
Sorry for the late reply. I've pinpointed the issue in the file chat_agent.py, accessible here: https://github.com/camel-ai/camel/blob/master/camel/agents/chat_agent.py#L929 When using the OAI provided by Qwen, setting the model_backend to n=3 results in only one choice in the response. A simple fix would be to verify if the number of choices matches the expected count of n. If it doesn't, you can re-run self.model_backend.run(openai_messages) to get more results. Since chat_agent is a crucial component, I'm not sure if this is the best solution. |
9 tasks
13 tasks
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Required prerequisites
Motivation
raised by @coolbeevip https://github.com/orgs/camel-ai/discussions/985
In the Role-Playing critic scenario, the user agent receives multiple response options through the n parameter of the OpenAI API, from which the critic selects one to guide the assistant agent. (If my understanding is incorrect, please feel free to point it out!)
However, this mechanism only works well with the OpenAI API; a brief investigation shows that other models do not support the n parameter. Should we ensure consistency in this behavior when using non-OpenAI models through engineering methods? (For example, by looping n times and concatenating the results.)
Solution
No response
Alternatives
No response
Additional context
No response
The text was updated successfully, but these errors were encountered: