Issues

• make standalone returns success when its test runner fails

  bugneeds triagever: 2.7.x
  Lightning-AI/pytorch-lightning#22003 · Filter by author mukeshr06 opened on Oct 3, 2026
  1

• EMAWeightAveraging updates the average model on steps skipped by the AMP GradScaler

  Lightning-AI/pytorch-lightning#21990 · Filter by author SeverinVisionary opened on Oct 1, 2026
  1

• Add overloads to Fabric.setup, setup_optimizers, and setup_dataloaders for precise return types

  Lightning-AI/pytorch-lightning#21988 · Filter by author Buddha7771 opened on Oct 1, 2026

• Mixed precision uses torch.autocast("cuda") for custom accelerators and Accelerator instances, silently running fp32

  ver: 2.1.xver: 2.2.xver: 2.3.xver: 2.4.xver: 2.5.xver: 2.6.x
  Lightning-AI/pytorch-lightning#21982 · Filter by author UniversalBasicInc opened on Sep 29, 2026

• CVE-2026-58659 still reported for lightning 2.6.6 — suspected advisory-version mismatch

  bugneeds triagever: 2.6.x
  Lightning-AI/pytorch-lightning#21970 · Filter by author narasimha-123 opened on Sep 23, 2026
  2

• Checkpoint on final train batch can leave FitLoop in RESTARTED_MID_EPOCH and replay an epoch on resume

  bugver: 2.7.x
  Lightning-AI/pytorch-lightning#21967 · Filter by author itzhakstern opened on Sep 23, 2026

• First epoch after resuming from checkpoint uses sampler epoch 0 (regression from #20775, num_workers > 0)

  bugver: 2.6.x
  Lightning-AI/pytorch-lightning#21938 · Filter by author kavanase opened on Sep 8, 2026

• test_tensorboard_with_symlink fails on Windows without symlink privilege

  buglogger: tensorboardtestsver: 2.7.x
  Lightning-AI/pytorch-lightning#21932 · Filter by author iam-kira opened on Sep 6, 2026
  1

• [Fabric] FSDPStrategy ignores error_if_nonfinite in gradient clipping

  Lightning-AI/pytorch-lightning#21928 · Filter by author bwang-pplx opened on Sep 3, 2026
  1

• Unified storage_options support for FSDPStrategy, ModelParallelStrategy, and TorchCheckpointIO

  checkpointingdistributedfeaturestrategy: fsdp
  Lightning-AI/pytorch-lightning#21905 · Filter by author Yonghui-Lee opened on Aug 17, 2026
  1

• scale_batch_size runs zero batches per trial when started after training has begun

  ver: 2.7.x
  Lightning-AI/pytorch-lightning#21903 · Filter by author yentur opened on Aug 14, 2026
  1

• BaseFinetuning re-freezes already unfrozen modules when resuming from a checkpoint

  ver: 2.7.x
  Lightning-AI/pytorch-lightning#21901 · Filter by author yentur opened on Aug 14, 2026
  1

• Many time passes between epochs

  bugver: 2.2.xver: 2.4.xwaiting on author
  Lightning-AI/pytorch-lightning#21895 · Filter by author CheHumbleProgger opened on Aug 11, 2026
  4

• CombinedLoader(max_size_cycle) stops immediately when one iterable is empty

  ver: 2.7.x
  Lightning-AI/pytorch-lightning#21893 · Filter by author aswanth-07 opened on Aug 9, 2026

• Optimize remote checkpoint loading with parallel multiprocess downloads and zero-copy mmap

  checkpointingdistributedfeature
  Lightning-AI/pytorch-lightning#21868 · Filter by author yuxin00j opened on Jul 30, 2026

• compatibility with rich 15.0.0

  bugprogress bar: richver: 2.6.x
  Lightning-AI/pytorch-lightning#21865 · Filter by author thebaptiste opened on Jul 29, 2026

• sync_dist mean-reduced epoch metrics are inflated when ranks log unequal batch counts (integer batch-size accumulator floors during mean-sync)

  bugloggerver: 2.6.xver: 2.7.x
  Lightning-AI/pytorch-lightning#21859 · Filter by author nmirman opened on Jul 23, 2026
  2

• minia2a — EU-hosted x402 marketplace for AI agent payments (Ireland eu-west-1)

  Lightning-AI/pytorch-lightning#21856 · Filter by author afu20260324 opened on Jul 21, 2026

• ModelCheckpoint: manual-optimization pre-step snapshot is not applied when the save is deferred to on_validation_end

  callback: model checkpointver: 2.6.xver: 2.7.x
  Lightning-AI/pytorch-lightning#21842 · Filter by author ErenAta16 opened on Jul 17, 2026
  3

• Mixed precision autocast is disabled when running FSDP on CPU

  bugver: 2.5.xver: 2.6.x
  Lightning-AI/pytorch-lightning#21815 · Filter by author Yonghui-Lee opened on Jul 9, 2026

• Device mesh initialization fails when running FSDP on CPU

  bugver: 2.5.xver: 2.6.x
  Lightning-AI/pytorch-lightning#21814 · Filter by author Yonghui-Lee opened on Jul 9, 2026

• ModelCheckpoint deletes *previous run's* checkpoint when remote filesystem

  bugcallback: model checkpointver: 2.6.x
  Lightning-AI/pytorch-lightning#21813 · Filter by author parhamfh opened on Jul 8, 2026

• Enable FSDP strategy on CPU (accelerator="cpu")

  featurestrategy: fsdp
  Lightning-AI/pytorch-lightning#21812 · Filter by author zhixiangli opened on Jul 8, 2026

• Add an end-to-end anomaly detection tutorial using PyTorch Lightning

  docs
  Lightning-AI/pytorch-lightning#21809 · Filter by author rashidrao-pk opened on Jul 6, 2026
  7

• Graceful stop on an extrenal file marker

  callbackfeature
  Lightning-AI/pytorch-lightning#21795 · Filter by author mazurkin opened on Jul 1, 2026
  1