Kernels
optimizer / torch-ext

Commit History

  1. fix(optimizer): resolve bug where weight decay was multiplied by wrong lr value

    Verified signature
    3356827

  1. Support HSDP (#4)⁠

    Verified signature
    8447fd1

  1. fix(muon): handle un-distributed env

    1f13dae
  2. refactor(muon): change argument adam_wd to weight_decay and handle params' type

    02ac540

  1. fix(muon): free tensors that are no longer needed

    64757cb
  2. chore(muon): update comment

    036642a
  3. chore(muon): clean build and update doc

    febdf5b

  1. fix(muon): delete intermediate tensors immediately to lower peak mem usage

    bdd2678
  2. chore: initial commit

    8535e80