-
Notifications
You must be signed in to change notification settings - Fork 707
Feat: Add trace_agg_mode to support both transition and trajectory aggregation
#134
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
base: main
Are you sure you want to change the base?
Conversation
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.
Pull Request Overview
This PR adds a trace_agg_mode configuration parameter to support both "transition" and "trajectory" level trace aggregation, with "transition" as the default for backward compatibility.
- Introduces a
trace_agg_modeparameter that controls aggregation behavior - Implements trajectory-level aggregation with response masking for multi-turn conversations
- Adds configuration option to the YAML config file
Reviewed Changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.
| File | Description |
|---|---|
| agentlightning/verl/trainer.py | Passes trace_agg_mode config to daemon and adds debugging breakpoint |
| agentlightning/verl/daemon.py | Implements trajectory aggregation logic and response mask generation |
| agentlightning/verl/config.yaml | Adds trace_agg_mode configuration option with default value |
Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.
fa72297 to
0d2dece
Compare
Add a
trace_agg_modeoption to control whether the trace aggregation is performed per transition or per trajectory (by adding response mask for multi-turn). Default remains 'transition' for backward compatibility.