All pull requests

Pull requests

• FSDP: Added fix + Regression test for incorrect params_group mapping in optimizer state sharding

  CLA Signed
  #1202·Shasha-17 opened Dec 26, 2025·6 tasks·

  ·
  2

• Fix failure in test_checkpoint_activations_norm.py

  CLA Signed
  #1201·maksverver opened Sep 13, 2025·12 tasks·

  ·

• Fix test failures caused by warns(None).

  CLA Signed
  #1200·maksverver opened Sep 13, 2025·12 tasks·

  ·

• Add support for free-threaded CPython 3.13t

  CLA Signed
  #1198·andfoy opened May 5, 2025·2 of 10 tasks·

  ·
  1

• support for grad acc

  CLA Signed
  #1190·ngoyal2707 opened Aug 30, 2024Contributor·

  ·
  2

• Bump scikit-learn from 1.1.3 to 1.5.0

  CLA Signeddependencies
  #1187·dependabot[bot] opened Jun 18, 2024Bot·

  ·

• FP8 AllGather Support in Fairscale

  CLA Signed
  #1185·levendlee opened May 20, 2024Member·10 tasks·

  ·
  6

• Llama4 FP8 Training Debug - fairscale

  CLA Signed
  #1183·jiecaoyu opened May 7, 2024·10 tasks·

  ·

• Fix minor grammatical corrections in docs

  CLA Signed
  #1181·aakashapoorv opened Apr 29, 2024·6 of 10 tasks·

  ·

• [FSDPv1] Only perform cat() during last microbatch backward() within FlattenParamsWrapper

  CLA Signed
  #1180·chrisxcai opened Apr 29, 2024·

  ·
  5

• [WIP] Make FSDPv1 only perform cat() during last microbatch backward() within FlattenParamsWrapper

  CLA Signed
  #1178·chrisxcai opened Apr 19, 2024·

  ·
  2

• Make sure that tensor is contiguous before gathering across processes

  CLA Signed
  #1173·patrickvonplaten opened Apr 1, 2024·

  ·

• Bump black from 22.3.0 to 24.3.0

  CLA Signeddependencies
  #1169·dependabot[bot] opened Mar 20, 2024Bot·

  ·

• Fairscale support for only performing allreduce in last microbatch

  CLA Signed
  #1168·jiecaoyu opened Mar 18, 2024·10 tasks·

  ·
  3

• Fix params_with_grad in FSDP when the model has frozen parameters

  CLA Signed
  #1167·whbldhwj opened Mar 4, 2024·1 of 10 tasks·

  ·

• Avoid calling _free_fp16_param_shard() too early

  CLA Signed
  #1164·jiecaoyu opened Feb 20, 2024·10 tasks·

  ·
  3

• Add support for torch.set_default_device when initializing model parameters

  CLA Signed
  #1158·fshp971 opened Dec 28, 2023·6 of 10 tasks·

  ·
  1

• Fix _free_full_params()

  CLA Signed
  #1149·hadasah opened Nov 21, 2023·10 tasks·

  ·

• Extend CheckpointFunction to track all tensor input/output

  CLA Signed
  #1148·000Justin000 opened Nov 7, 2023·6 of 10 tasks·

  ·
  1

• [Not for merge] fp8allgather debug

  CLA Signed
  #1147·jiecaoyu opened Oct 29, 2023·

  ·

• Add main grad before fwd pass

  CLA Signed
  #1142·vedanuj opened Oct 4, 2023·

  ·
  10

• Add main_grad

  CLA Signed
  #1140·jianyuh opened Oct 2, 2023Member·10 tasks·

  ·
  6

• Fp8 all gather hack

  CLA Signed
  #1136·jspark1105 opened Sep 17, 2023·

  ·
  6

• Fix typo in ParallelEmbedding argument assignment

  CLA Signed
  #1132·hessamb opened Aug 28, 2023·

  ·
  4

• Add a context manager for activation sharding.

  CLA Signed
  #1127·luyug opened Jun 12, 2023·

  ·
  1