Pythagora-io / gpt-pilot

The first real AI developer
Other
29.06k stars 2.91k forks source link

[Enhancement]: v0.2 doesn't give the 'retry' option anymore #990

Closed niklasfink closed 2 weeks ago

niklasfink commented 4 weeks ago

Version

VisualStudio Code extension

Suggestion

Unfortunately, the v0.2 doesn't have the retry option anymore, which could be used if the request to the LLM failed. A request to the LLM can temporary fail due to various reasons, so it made a lot of sense to be able to just repeat the action. In the current version, Pythagora just hangs saying "Waiting..." after the request has failed and you need to restart it completely.

image

senko commented 4 weeks ago

Hey @niklasfink thanks for reporting this!

Could you post a screenshot showing the complete stack trace (feel free to mask out the directory/username bits for privacy).

We've added manual retries (and fixed some of the missing error catches) in https://github.com/Pythagora-io/gpt-pilot/pull/973 and I'd like to check if we've missed some edge case or if this is just due to the changes being in main but not released yet.

niklasfink commented 4 weeks ago

Hi @senko, the error shown is just the error output of the LLM API. In this case, I'm using litellm and it had a bug with long lasting responses. However, Pythagora/gpt-pilot also fails to detect the error when I cancel the connection for testing or when another LLM error ocurrs. In the previous version the error detection was reliable and I was able to easily repeat the previous LLM call in all of my cases.

image

niklasfink commented 2 weeks ago

Works better now with 0.2.2+