-
Notifications
You must be signed in to change notification settings - Fork 5.3k
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
Expanded speaker name matching during speaker selection #2222
Conversation
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## main #2222 +/- ##
===========================================
+ Coverage 37.83% 49.71% +11.88%
===========================================
Files 77 77
Lines 7766 7766
Branches 1663 1800 +137
===========================================
+ Hits 2938 3861 +923
+ Misses 4579 3583 -996
- Partials 249 322 +73
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
Let me know if you want me to create a new PR with a branch in the AutoGen repository for testing, rather than my own. |
Yes, please do that and resolve the conflict. |
No problem, all done and this can be closed. New PR: #2267 |
Why are these changes needed?
During the speaker selection process, the LLM returns the name of the next speaker. This is fairly reliable with OpenAI's models but with open-source/weight models they struggle with returning just the single agent name and sticking to the format of the name.
In particular two issues are:
In the proposed changes:
GroupChat
's_mentioned_agents
function, agent names will now match under any of the following conditions (all continue to be case-sensitive):\_
(note: single backslash) instead of_
(e.g. 'Story_writer' == 'Story\_writer')I've added a test function to
test_groupchat.py
.If we proceed with this, I'd like to add to the tips for non-OpenAI models documentation as part of this PR.
Please see #1746 which outlines a number of issues with open-source/weight models and this is a part of addressing them. I don't believe this will affect the speaker selection process when using OpenAI's models.
My plan is to continue to create PRs to address these shortcomings and then close off #1746. If it is preferred that I consolidate them into one PR, please let me know.
Related issue number
Based on shortcomings identified in #1746.
Checks