All pull requests

Pull requests

• Resume completed epochs normally when max_steps is set

  #21999·mukeshr06 opened Oct 3, 2026·5 tasks done·

  ·
  1

• Size Rich metric columns by terminal display width

  #21998·mukeshr06 opened Oct 3, 2026·4 tasks done·

  ·
  1

• Honor explicit repeated PyTorch profiler schedules

  #21997·mukeshr06 opened Oct 2, 2026·4 tasks done·

  ·
  1

• Preserve async checkpoint save/remove ordering

  #21996·mukeshr06 opened Oct 2, 2026·4 tasks done·

  ·
  1

• Make MNIST helper tests independent of downloads

  #21995·mukeshr06 opened Oct 2, 2026·3 tasks done·

  ·

• Fix inflated tqdm rates after resuming mid-epoch

  #21994·mukeshr06 opened Oct 2, 2026·3 tasks done·

  ·
  1

• Fix CLI config saving without a local logger directory

  #21993·aswanth-07 opened Oct 2, 2026·4 of 7 tasks·

  ·
  1

• Fix EMA updates after skipped AMP optimizer steps

  #21991·anishhongekar47-dev opened Oct 1, 2026·4 of 12 tasks·

  ·
  1

• docs: fix typo in trainer.rst

  #21989·adosar opened Oct 1, 2026Contributor·7 tasks·

  ·

• fix: AMP autocast device for accelerator instances and custom registered accelerators

  #21986·deependujha opened Sep 30, 2026Collaborator·7 tasks·

  ·
  1

  2

• Fix FitLoop leaving _restart_stage in RESTARTED_MID_EPOCH after fast-forwarding an epoch

  #21985·mayuriphad opened Sep 30, 2026·

  ·
  1

  1

• Fix ckpt_path="last" not finding last checkpoints with a templated CHECKPOINT_NAME_LAST

  #21984·breken-ai opened Sep 30, 2026·3 of 7 tasks·

  ·

• Fix CSVLogger automatic step restarting at 0 after every flush

  #21983·breken-ai opened Sep 30, 2026·3 of 7 tasks·

  ·

• fix: preserve stateful dataloader state around estimated_stepping_batches

  #21980·deependujha opened Sep 29, 2026Collaborator·7 tasks·

  ·
  1

  11

• fix: set the sampler epoch before the training iterator is created on resume

  #21979·deependujha opened Sep 29, 2026Collaborator·7 tasks·

  ·
  1

  2

• CI: update transformer engine & 1-bit adam DeepSpeed links in documentation and code comments

  #21978·deependujha opened Sep 29, 2026Collaborator·7 tasks·

  ·
  3

• docs: fix CodeFactor badge and 1-bit Adam links

  #21977·pratikgx opened Sep 28, 2026·

  ·

• Defer logger flushing to logger policies

  #21976·rohan-patnaik opened Sep 27, 2026·

  ·
  1

• feat: add enable_device_summary flag to Trainer (fixes #13378)

  #21975·CJstate opened Sep 27, 2026·

  ·
  1

• docs: add :emphasize-lines: to code blocks in warnings.rst (fixes #12119)

  #21974·CJstate opened Sep 27, 2026·

  ·
  1

• Fix trainer.current_epoch not restored when validating/testing/predicting with ckpt_path

  #21971·ankurdhuriya opened Sep 26, 2026·5 tasks done·

  ·
  1

• Fix FitLoop restart state after last batch checkpoint

  #21969·itzhakstern opened Sep 23, 2026Contributor·

  ·
  1

  4

• Start a new epoch after fast-forwarding one restored from its last batch

  #21968·dchaudhari7177 opened Sep 23, 2026·3 tasks done·

  ·
  1

  3

• Save training topology metadata in full checkpoints

  #21966·mukeshr06 opened Sep 23, 2026·

  ·
  1

• Fix extract_batch_size crash on NumPy ndarray batches

  #21965·MohammadHijjawi97 opened Sep 22, 2026·

  ·