All-Hands-AI / OpenHands

🙌 OpenHands: Code Less, Make More
https://all-hands.dev
MIT License
31.55k stars 3.65k forks source link

Pause and Restart controls are often unresponsive #1427

Open arose26 opened 4 months ago

arose26 commented 4 months ago

What problem or use case are you trying to solve? Many times the AI will get stuck trying to do a futile task. The user must currently wait for it to finish. Describe the UX of the solution you'd like An ABORT button that cancels the current task. Do you have thoughts on the technical implementation? Kill current process and return input to user.

zhonggegege commented 4 months ago

Are you not used to the "MonologueAgent" monologue mode? A code reply that is easy to complete in the dialogue mode is difficult to complete after dozens of rounds in the monologue mode. It always has various errors, such as repeating without action in the monologue mode, Problems such as writing only part of the code, repeating multiple pieces of the same code, etc. In this case, using the chat mode can directly reply to the complete code in 2-5 rounds of conversations. Scenarios like this feel like simple things become complicated. Yes, I think there is a problem with the logic. Of course, I don’t mean that the monologue logic is unnecessary, but I think it still has a lot of room for optimization.

enyst commented 4 months ago

@arose26 We have added a 'pause' button and a 'restart a new agent task' button under the chat box in the UI. Do those address this issue? You may need to upgrade, it's fairly recent I believe.

@zhonggegege You may see better results with state of the art models (like GPT-4, Claude-3), but this is indeed work in progress. As far as we know, monologue agent is still the most used, SWE agent might be better on some specific actions, although that's not enough to do well with GPT-3.5, but again it seems better with GPT-4. There is a lot of work on CodeActAgent with conversation mode, that you may see results of, either now or soon. You can also take a look at the current PRs on CodeAct and try them out if you wish.

zhonggegege commented 4 months ago

When using local agent LLM, when the agent starts running, in many cases the "Pause" and "Restart" buttons are invalid (no response when clicked). If you observe "Terminal", "WEB-UI", and "LLM Proxy Server (Response Information)" at the same time, you will feel that the response of the front-end WEB-UI is very uncoordinated and cannot respond in time. It feels like there is a huge delay. The picture suddenly returned to normal.

rbren commented 4 months ago

Agree--the controls often become unresponsive. I think that's the bigger issue here, so I've changed the title to reflect it

github-actions[bot] commented 3 months ago

This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.

SmartManoj commented 2 months ago

@tobitege, #2602 async feature will solve this issue. right?

tobitege commented 2 months ago

Yes.

github-actions[bot] commented 1 month ago

This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.