Skip to content

[Zephyr] Send task stats from Zephyr to Iris#1

Closed
wmoss wants to merge 1 commit intomainfrom
04-23-_zephyr_send_task_stats_from_zephyr_to_iris
Closed

[Zephyr] Send task stats from Zephyr to Iris#1
wmoss wants to merge 1 commit intomainfrom
04-23-_zephyr_send_task_stats_from_zephyr_to_iris

Conversation

@wmoss
Copy link
Copy Markdown
Owner

@wmoss wmoss commented Apr 23, 2026

Describe what changed and why. Follow with bullets for
specific changes if needed. Keep it concise — this text becomes the squash-merge
commit message, so avoid markdown formatting (headers, tables, images).

e.g.

Title: [RL] Fix loss: use global token normalization instead of per-example

"""
This fixes a regression in the DAPO loss computation by switching
from per-example normalization (/ n_i) back to global token
normalization (/ N). Per-example normalization gives shorter responses
disproportionately more gradient weight, which hurts math reasoning
tasks where correct answers often require detailed, longer derivations.
Global normalization weights all examples equally regardless of response
length.
"""

Fixes #

Takes the stats that were added in marin-community#5063 and reports them to the Iris controller. For now, we're just logging the stats from Iris, but subsequent changes will store and display them.
Copy link
Copy Markdown
Owner Author

wmoss commented Apr 23, 2026

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.

1 participant