feat: add termination parameter to ChatAgent step and astep #2285
Conversation
camel/agents/chat_agent.py
Outdated
| self, | ||
| input_message: Union[BaseMessage, str], | ||
| response_format: Optional[Type[BaseModel]] = None, | ||
| stop_event: Optional[threading.Event] = None, |
There was a problem hiding this comment.
thanks @a7m-1st ! Would it better to add this argument in __init__ instead of step?
There was a problem hiding this comment.
Definitely would make the code cleaner! Let me do that.
Thanks @Wendong-Fan .
| extend_task_specify_meta_dict: Optional[Dict] = None, | ||
| output_language: Optional[str] = None, | ||
| stop_event: Optional[threading.Event] = None, | ||
| ) -> None: | ||
| if model is not None: |
There was a problem hiding this comment.
Hi there, @Wendong-Fan as suggested I have elevated the stop_event to be part of ChatAgent, thus passing the event turned to:
run_society -> society.stop_event = stop_event -> which is passed to ChatAgent in
__init__
Since society:OwlRolePlaying has stop_event now, this resulted in updating OwlRolePlaying in camel-ai/owl#450 which is reflected to its parent class RolePlaying here.
There was a problem hiding this comment.
Also, if updating RolePlaying is not appropriate We can revert the last commit ae419f2 and it would constrain changes to OwlRolePlaying only.
| extend_task_specify_meta_dict: Optional[Dict] = None, | ||
| output_language: Optional[str] = None, | ||
| stop_event: Optional[threading.Event] = None, | ||
| ) -> None: | ||
| if model is not None: |
| return self._step_token_exceed( | ||
| num_tokens, tool_call_records, "termination_triggered" | ||
| ) |
There was a problem hiding this comment.
_step_token_exceed is a bit confusing since the agent is not terminated since the number of token exceeding the context windows. Could we improve this?
There was a problem hiding this comment.
Oh yes thanks I forgot to mention.
@lightaime Since the function is generic, how about we rename it to _step_terminate because it terminates then returns token number and called functions?
Description
Terminating an agent safely as of now was not possible. That was due to the one-shot nature of the tasks. But when rerunning is required in OWL for instance in camel-ai/owl#450, safe and quick termination is of utmost importance.
This branch is last of three PRs, this one adds a termination event in
chat_agent.pywhich allows atomic termination and patches a runtime error (empty msg) raised byagent_response.pywhen terminating Agents.The other PR is #2194, which fixes issue camel-ai/owl#449 and introduces
terminate_browsertool to browser toolkit.Lastly the PR in owl camel-ai/owl#450, uses the terminate_browser tool and passes the stop_event to the ChatAgent.
Checklist
Go over all the following points, and put an
xin all the boxes that apply.Fixes #issue-numberin the PR description (required)pyproject.tomlanduv lockIf you are unsure about any of these, don't hesitate to ask. We are here to help!