Issues

• save_last="link": a crash while relinking last.ckpt makes ckpt_path="last" train from scratch

  Lightning-AI/pytorch-lightning#22179 · Filter by author asonnino opened on Nov 13, 2026
  1

• Failed validation, test, or prediction leaves model submodules in evaluation mode

  bugneeds triagever: 2.7.x
  Lightning-AI/pytorch-lightning#22176 · Filter by author Afloat16 opened on Nov 13, 2026

• LearningRateMonitor history aliases tensor learning rates updated by schedulers

  bugneeds triagever: 2.7.x
  Lightning-AI/pytorch-lightning#22174 · Filter by author Afloat16 opened on Nov 13, 2026

• self.log(enable_graph=True) loses the tensor autograd graph

  bugneeds triagever: 2.7.x
  Lightning-AI/pytorch-lightning#22172 · Filter by author Afloat16 opened on Nov 13, 2026

• LR finder omits optimizer updates at partial accumulation boundaries

  bugneeds triagever: 2.7.x
  Lightning-AI/pytorch-lightning#22170 · Filter by author Afloat16 opened on Nov 13, 2026

• Batch-size finder can ignore a smaller init_val and reuse the pre-search loader

  bugneeds triagever: 2.7.x
  Lightning-AI/pytorch-lightning#22168 · Filter by author Afloat16 opened on Nov 13, 2026

• LightningOptimizer.toggle_model leaves other parameters frozen after an exception

  bugneeds triagever: 2.7.x
  Lightning-AI/pytorch-lightning#22164 · Filter by author Afloat16 opened on Nov 12, 2026

• ModelPruning checkpoint saving mistakes causal attention buffers for pruning masks

  bugneeds triagever: 2.7.x
  Lightning-AI/pytorch-lightning#22162 · Filter by author Afloat16 opened on Nov 12, 2026

• CombinedLoader ignores per-loader limits during non-sequential iteration

  Lightning-AI/pytorch-lightning#22160 · Filter by author Afloat16 opened on Nov 12, 2026

• BaseFinetuning unfreezes generator modules without adding their parameters to the optimizer

  bugneeds triagever: 2.7.x
  Lightning-AI/pytorch-lightning#22158 · Filter by author Afloat16 opened on Nov 12, 2026

• WeightAveraging loses training weights in checkpoints saved at normal fit completion

  bugneeds triagever: 2.7.x
  Lightning-AI/pytorch-lightning#22155 · Filter by author Afloat16 opened on Nov 12, 2026

• Let a callback stop training before the current batch runs

  Lightning-AI/pytorch-lightning#22151 · Filter by author rfilgueiras opened on Nov 11, 2026

• Avoid redundant parameter mutations when toggling optimizers

  Lightning-AI/pytorch-lightning#22143 · Filter by author Hang-get opened on Nov 10, 2026

• [Feature Request] support Ascend NPU as a backend accelerators

  featureneeds triage
  Lightning-AI/pytorch-lightning#22136 · Filter by author zhoufei-tech opened on Nov 9, 2026

• CPU ddp_spawn workers SIGABRT at exit: _init_dist_connection only destroys the process group for NCCL

  ver: 2.7.x
  Lightning-AI/pytorch-lightning#22124 · Filter by author glaziermag opened on Nov 5, 2026
  1

• Support TorchTPU (torch_tpu) accelerator in Fabric and PyTorch Lightning

  featureneeds triage
  Lightning-AI/pytorch-lightning#22119 · Filter by author zhixiangli opened on Nov 4, 2026

• GradientAccumulationScheduler rejects its own previous assignment on a second fit

  Lightning-AI/pytorch-lightning#22102 · Filter by author anishpentyala opened on Nov 1, 2026
  1

• Repeated compiled-model Trainer fits leave Dynamo wrappers active after to_uncompiled

  Lightning-AI/pytorch-lightning#22101 · Filter by author anishpentyala opened on Nov 1, 2026

• WeightAveraging validation exceptions reverse live/average weights in exception checkpoints

  Lightning-AI/pytorch-lightning#22100 · Filter by author anishpentyala opened on Nov 1, 2026

• Reused WeightAveraging skips fresh-fit updates and restores initial weights

  Lightning-AI/pytorch-lightning#22099 · Filter by author anishpentyala opened on Nov 1, 2026

• ModelPruning lottery-ticket reset aliases the saved initial weights on CPU

  Lightning-AI/pytorch-lightning#22098 · Filter by author anishpentyala opened on Nov 1, 2026

• ThroughputMonitor retains the previous model FLOPs capability when reused

  Lightning-AI/pytorch-lightning#22097 · Filter by author anishpentyala opened on Nov 1, 2026

• LR finder leaves temporary callbacks and search limits after a model exception

  Lightning-AI/pytorch-lightning#22096 · Filter by author anishpentyala opened on Nov 1, 2026

• AsyncCheckpointIO skips the wrapped backend teardown hook

  Lightning-AI/pytorch-lightning#22095 · Filter by author anishpentyala opened on Nov 1, 2026

• AsyncCheckpointIO retains live NumPy arrays while snapshotting model tensors

  Lightning-AI/pytorch-lightning#22086 · Filter by author salma2vec opened on Oct 31, 2026