-
Notifications
You must be signed in to change notification settings - Fork 44.5k
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
Update prompt.py to clarify how to call commands #4027
Conversation
The latest updates on your projects. Learn more about Vercel for Git ↗︎ 1 Ignored Deployment
|
lgtm! |
#RichTheLightning |
nice, I've seen code reviews here where people pointed out discrepancies in that area. the command string should probably be internally validated at the command mgr level using a regex to show a warning, especially since commands are likely to be added by external means, like plugins. See also: |
This PR exceeds the recommended size of 200 lines. Please make sure you are NOT addressing multiple issues with one PR. Note this PR might be rejected due to its size |
1 similar comment
This PR exceeds the recommended size of 200 lines. Please make sure you are NOT addressing multiple issues with one PR. Note this PR might be rejected due to its size |
This PR exceeds the recommended size of 200 lines. Please make sure you are NOT addressing multiple issues with one PR. Note this PR might be rejected due to its size |
This PR exceeds the recommended size of 200 lines. Please make sure you are NOT addressing multiple issues with one PR. Note this PR might be rejected due to its size |
Codecov ReportPatch and project coverage have no change.
Additional details and impacted files@@ Coverage Diff @@
## master #4027 +/- ##
=======================================
Coverage 60.72% 60.72%
=======================================
Files 73 73
Lines 3320 3320
Branches 475 475
=======================================
Hits 2016 2016
Misses 1164 1164
Partials 140 140
☔ View full report in Codecov by Sentry. |
This PR exceeds the recommended size of 200 lines. Please make sure you are NOT addressing multiple issues with one PR. Note this PR might be rejected due to its size |
This PR exceeds the recommended size of 200 lines. Please make sure you are NOT addressing multiple issues with one PR. Note this PR might be rejected due to its size |
This PR exceeds the recommended size of 200 lines. Please make sure you are NOT addressing multiple issues with one PR. Note this PR might be rejected due to its size |
This PR exceeds the recommended size of 200 lines. Please make sure you are NOT addressing multiple issues with one PR. Note this PR might be rejected due to its size |
FYI: I've seen fewer command hallucinations by adapting the prompt generator to encourage the LLM to respond within the bounds of the list of commands by also adding the length of the list to the prompt itself (aka "pick one between 1-21"). |
…s#4027) * Update prompt.py to clarify how to call commands * new cassettes * try again with a new cassette * one last try with the cassette * black dot --------- Co-authored-by: k-boikov <[email protected]> Co-authored-by: Nicholas Tindle <[email protected]>
Background
As discussed on the call recently, we tell the LLM to exclusively use the commands in double quotes, but the commands aren't listed in double quotes, and are underscore-separated and not space-separated.
Changes
Documentation
Test Plan
PR Quality Checklist