Skip to content

Conversation

rakkit
Copy link
Contributor

@rakkit rakkit commented Aug 25, 2025

This is a distributed version of Scion or Modular Norm, muon is considered to be a variant of this by using explicit AdamW for LLM's embedding/output.

Works:

  • Embedding/head
  • FSDP/DP/TP/EP/CP/PP parameters
  • Bias (mainly for norm)
  • weight decay

Missing

  • Conv

Need some extra work to adjust the EP changes for EP-[shard(1)] and ETP?

At the moment, we need a long configuration to initialize it, will fix later by passing the list

CC @janEbert @ofivite

@meta-cla meta-cla bot added the CLA Signed This label is managed by the Meta Open Source bot. label Aug 25, 2025
Copy link
Contributor

@tianyu-l tianyu-l left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for the PR on cutting-edge features!

I didn't read the papers so please forgive me if what I comment doesn't make sense.

I guess for "core" changes such as this one on optimizers, the recommended path is to first land in pytorch/pytorch, and then expose minimal interfaces to torchtitan. torchtitan shouldn't be a place to host core features.

cc @janeyx99 on interesting optimizer work

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
CLA Signed This label is managed by the Meta Open Source bot.
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants