Skip to content

docs: clarify VERL agent-mode config compatibility - #536

Open
Valor (zrh805) wants to merge 2 commits into
microsoft:mainfrom
zrh805:docs/clarify-verl-config
Open

docs: clarify VERL agent-mode config compatibility#536
Valor (zrh805) wants to merge 2 commits into
microsoft:mainfrom
zrh805:docs/clarify-verl-config

Conversation

@zrh805

Copy link
Copy Markdown

Summary

  • document the agent-mode behavior of temperature, top_p, val_kwargs, and data.truncation
  • distinguish standard VERL settings from the paths Agent Lightning replaces
  • link readers to the version-matched upstream VERL configuration reference

Fixes #464

Validation

  • git diff --check
  • uv tool run --from mkdocs-material --with mkdocs-git-revision-date-localized-plugin --with mkdocs-git-authors-plugin --with mkdocs-macros-plugin --with 'mkdocstrings[python]' --with mike --with mkdocs-autorefs mkdocs build --strict\n\nThe repository lockfile only supports Linux, so its native uv sync --group dev cannot resolve on macOS; the strict build above used an isolated environment with the plugins declared in mkdocs.yml.

Copilot AI review requested due to automatic review settings August 2, 2026 16:46

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

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 updates the VERL algorithm documentation to clarify which upstream VERL configuration fields are effective (or overridden) when running Agent Lightning in agent-mode, aligning expectations for users configuring agl.VERL(...).

Changes:

  • Adds a “Configuration compatibility” section describing how Agent Lightning composes VERL configs and which settings behave differently in agent-mode.
  • Documents the agent-mode handling of temperature, rollout sampling parameters (e.g., top_p), val_kwargs, and data.truncation.
  • Links to the upstream VERL configuration reference for further configuration details.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Comment thread docs/algorithm-zoo/verl.md Outdated
Comment on lines +63 to +66
Other settings, including optimizer, model, worker, checkpoint, and scheduling
options, are still interpreted by the installed VERL version. Check the
[VERL configuration reference](https://verl.readthedocs.io/en/latest/examples/config.html)
for the version paired with your Agent Lightning installation.
@zrh805
Valor (zrh805) force-pushed the docs/clarify-verl-config branch from fe4259e to dc46384 Compare August 3, 2026 01:41
@zrh805

Copy link
Copy Markdown
Author

@microsoft-github-policy-service agree

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Which hyper-parameters in the standard VERL configuration do not take effect?

2 participants