Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Agent mode and omniscient mode #4

Open
XuhuiZhou opened this issue Aug 10, 2024 · 4 comments
Open

Agent mode and omniscient mode #4

XuhuiZhou opened this issue Aug 10, 2024 · 4 comments
Labels
enhancement New feature or request

Comments

@XuhuiZhou
Copy link
Member

The current chat mode for humans should be agent mode (we give them limited info and don't allow them to change things)

Let's have another chat mode we allow humans to see everything and all the info is editable.

@XuhuiZhou XuhuiZhou added the enhancement New feature or request label Aug 10, 2024
@bugsz
Copy link
Collaborator

bugsz commented Aug 12, 2024

Do we want the scenario to be editable in free-form? Or we provide some editable fields to ensure the background format is correct

@XuhuiZhou
Copy link
Member Author

I think for now let's just do the free form thing

@XuhuiZhou
Copy link
Member Author

Re your point in the slack,

class EpisodeLog(JsonModel):
    # Note that we did not validate the following constraints:
    # 1. The number of turns in messages and rewards should be the same or off by 1
    # 2. The agents in the messages are the same as the agetns

    environment: str = Field(index=True)
    agents: list[str] = Field(index=True)
    tag: str | None = Field(index=True)
    models: list[str] | None = Field(index=True)
    messages: list[list[tuple[str, str, str]]]  # Messages arranged by turn
    reasoning: str
    rewards: list[tuple[float, dict[str, float]] | float]  # Rewards arranged by turn
    rewards_prompt: str

what if we initiate a special pseudo pk for the user-created environment and agent? If we are not planning to store the user-created environment and agent?

@bugsz
Copy link
Collaborator

bugsz commented Aug 12, 2024

That makes sense but what if the user got disconnected? We have to handle it either.
After checking the render_for_human I found no primary key is needed, so that should not be a problem right now

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
enhancement New feature or request
Projects
None yet
Development

No branches or pull requests

2 participants