Skip to content

Conversation

@aaaandychen
Copy link

@aaaandychen aaaandychen commented Dec 29, 2025

this pr is related to issue: #314 , which is to extend metrics with 0 mean and unify post process of different backends.

  1. Added metric_process for metric reduction: Implemented backend data collection and formatted it into tensors for input. This method calculates mean values for specified parameters and provides extended metric support.
  2. Refactored log_process: Standardized the workflow to unify the metric reduction logic for both FSDP and Megatron.
    Testing confirms that FSDP maintains stable parameter magnitudes consistent with the baseline, while extending support for max, min, and std values for designated parameters.

chenzhenyang and others added 3 commits December 28, 2025 03:55
@gemini-code-assist
Copy link
Contributor

Important

Installation incomplete: to start using Gemini Code Assist, please ask the organization owner(s) to visit the Gemini Code Assist Admin Console and sign the Terms of Services.

@aaaandychen aaaandychen changed the title Improve logging process refactor: Improve logging process Dec 29, 2025
Signed-off-by: chenzhenyang <chenzhenyang@moonshot.cn>
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