11 Pull requests merged by 2 people

• build(deps): bump mypy from 1.11.0 to 1.15.0 in /requirements

  #20759 merged Apr 28, 2025

• drop mergify

  #20770 merged Apr 28, 2025

• docs: update repo link preventing flakiness

  #20769 merged Apr 28, 2025

• build(deps): update scikit-learn requirement from <1.4.0,>0.22.1 to >0.22.1,<1.7.0 in /requirements

  #20760 merged Apr 28, 2025

• build(deps): update pandas requirement from <2.2.0,>1.0 to >1.0,<2.3.0 in /requirements

  #20764 merged Apr 28, 2025

• build(deps): bump torch from 2.6.0 to 2.7.0 in /requirements

  #20765 merged Apr 28, 2025

• build(deps): update wheel requirement from <0.44.0 to <0.46.0 in /requirements

  #20763 merged Apr 28, 2025

• build(deps): bump pytest-cov from 4.1.0 to 6.1.1 in /requirements

  #20768 merged Apr 28, 2025

• build(deps): update requests requirement from <2.32.0 to <2.33.0 in /requirements

  #20762 merged Apr 28, 2025

• build(deps): update awscli requirement from <1.31.0,>=1.30.0 to >=1.30.0,<1.41.0 in /requirements

  #20766 merged Apr 28, 2025

• build(deps): update fsspec[http] requirement from <2024.4.0,>=2022.5.0 to >=2022.5.0,<2025.4.0 in /requirements

  #20767 merged Apr 28, 2025

8 Pull requests opened by 8 people

• add toggled_optimizer context manager

  #20771 opened Apr 28, 2025

• Fix double iteration bug when resumed from a checkpoint.

  #20775 opened Apr 29, 2025

• LightningCLI instantiator receives values applied by instantiation links to set in hparams

  #20777 opened Apr 30, 2025

• Added warmup parameter to early stopping cb

  #20778 opened Apr 30, 2025

• Allow cross-device local checkpoints with `fsspec>=2025.5.0`

  #20780 opened May 1, 2025

• check param is of nn.Parameter type

  #20783 opened May 3, 2025

• Support `grad_clip_norm_()` for FSDP

  #20784 opened May 3, 2025

• docs: update ref to latest tutorials

  #20785 opened May 4, 2025

1 Issue closed by 1 person

• Running LightningCLI from Python cannot pass a config.yaml file

  #20722 closed Apr 28, 2025

8 Issues opened by 8 people

• Proposal: GPU-safe utilities for Lightning (based on Universal CUDA Tools)

  #20782 opened May 3, 2025

• `ModelCheckpoint`'s argument `save_on_train_epoch_end`'s documentation unclear when value is `None`

  #20781 opened May 1, 2025

• Pytorch Profiler crashes while using it with Pytorch Lightning modules

  #20779 opened May 1, 2025

• slurm env incorrectly complains about srun with salloc interactive session.

  #20776 opened Apr 29, 2025

• WandbLogger: log distributed training experiments (multi-node)

  #20774 opened Apr 29, 2025

• `limit_val_batches` as sequence

  #20773 opened Apr 28, 2025

• Lightning is requiring packaging < 25.0

  #20772 opened Apr 28, 2025

• Can we have an LLM.txt?

  #20758 opened Apr 27, 2025

48 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• fix: `overfit_batches` uses same batch for train and val

  #20731 commented on Apr 28, 2025 • 3 new comments

• Integrate TorchTensorRt in order to increase speed during inference

  #11438 commented on May 4, 2025 • 0 new comments

• Add logger state dumping and restoring

  #13069 commented on Apr 29, 2025 • 0 new comments

• Lightning Module throws error when save_parameters() is not called from init

  #13240 commented on Apr 28, 2025 • 0 new comments

• Azure OpenMPI Environment

  #14118 commented on Apr 28, 2025 • 0 new comments

• [WIP] Support all CombinedLoader modes in prediction loop

  #17749 commented on Apr 28, 2025 • 0 new comments

• Support multi-run with hydra + DDP

  #18175 commented on Apr 28, 2025 • 0 new comments

• Update rich_progress.py

  #18505 commented on Apr 28, 2025 • 0 new comments

• feat(integrations): Improve checkpoint functionality of `WandbLogger`

  #19297 commented on Apr 29, 2025 • 0 new comments

• Enable support for Intel XPU devices (AKA Intel GPUs)

  #19443 commented on Apr 28, 2025 • 0 new comments

• [WIP] Basic system check for troubleshooting multi-GPU issues

  #19609 commented on Apr 28, 2025 • 0 new comments

• WIP: Integrate Collective into strategies

  #19881 commented on Apr 28, 2025 • 0 new comments

• modified num_replicas=self.world_size

  #19962 commented on Apr 28, 2025 • 0 new comments

• Test optimizer to device

  #20062 commented on Apr 28, 2025 • 0 new comments

• Feat: support reusable instance of `ModelCheckpoint`

  #20202 commented on Apr 29, 2025 • 0 new comments

• Support variable batch size in throughput callback

  #20236 commented on Apr 28, 2025 • 0 new comments

• Add compile_fn parameter for Trainer

  #20269 commented on Apr 28, 2025 • 0 new comments

• Add learning rate scheduling support for `DeepSpeedStrategy`

  #20320 commented on May 3, 2025 • 0 new comments

• Better registration support for a wide range of third-party hardware

  #20349 commented on Apr 27, 2025 • 0 new comments

• Add fp8 (torchao)/fsdp2/torch_compile handlers and tests

  #20445 commented on Apr 28, 2025 • 0 new comments

• fix(mlflow): Enabling multiple callbacks for checkpoint reporting

  #20585 commented on Apr 29, 2025 • 0 new comments

• feat[logger] update mlflow limit for parameters length log

  #20636 commented on Apr 29, 2025 • 0 new comments

• feat: add support for cloud files and non-local checkpoints

  #20701 commented on Apr 28, 2025 • 0 new comments

• updating issue template

  #20720 commented on Apr 28, 2025 • 0 new comments

• Stop after validation sanity checking.

  #20682 commented on Apr 28, 2025 • 0 new comments

• How to use Webdataset in DDP setting? ValueError: you need to add an explicit nodesplitter to your input pipeline for multi-node training

  #20149 commented on Apr 28, 2025 • 0 new comments

• self.all_gather does not work on on_train_epoch_end

  #20683 commented on Apr 28, 2025 • 0 new comments

• Add support for Comet experiment data flush after saving checkpoint into CometLogger

  #20681 commented on Apr 28, 2025 • 0 new comments

• Allow Using Specific Dataloaders in LightningCLI (Instead of a DataModule)

  #20673 commented on Apr 28, 2025 • 0 new comments

• CPU-Memory keeps accumulating during `trainer.predict`

  #19398 commented on Apr 28, 2025 • 0 new comments

• Non-reproducible results with num_workers=0

  #20679 commented on Apr 28, 2025 • 0 new comments

• RuntimeError when running basic GAN model (from tutorial at lightning.ai) with DDP

  #20328 commented on Apr 28, 2025 • 0 new comments

• Overfit batches parameter gives a validation batch

  #15021 commented on Apr 28, 2025 • 0 new comments

• warnings: resuming before epoch end is absolutely normal for long trainings

  #18780 commented on Apr 29, 2025 • 0 new comments

• Make sure the upcoming change in the default for `weights_only` from False to True is handled correctly

  #20450 commented on Apr 29, 2025 • 0 new comments

• `Trainer.validate()` after `Trainer.fit()` not working with FSDP and `auto_wrap_policy`

  #19015 commented on Apr 29, 2025 • 0 new comments

• `ModelCheckpoint` not saving best model

  #20657 commented on Apr 30, 2025 • 0 new comments

• `link_arguments` does not work in lightning 2.3

  #20147 commented on Apr 30, 2025 • 0 new comments

• DeepSpeed Stage 3 in lightning leads to Nan and Inf values in the model parameters.

  #20534 commented on Apr 30, 2025 • 0 new comments

• Make `save_hyperparameters` consistent for CLI and hardcoded training for custom python objects

  #20432 commented on Apr 30, 2025 • 0 new comments

• Incorrect `.hparams` when using `LightningCLI` with `Callable`

  #20726 commented on Apr 30, 2025 • 0 new comments

• `auto_lr_find` does not work if there is a BackboneFinetuning callback

  #14674 commented on Apr 30, 2025 • 0 new comments

• WandbLogger doesn't format config correctly

  #17558 commented on Apr 30, 2025 • 0 new comments

• Save save_hyperparameters no longer respects linked arguments.

  #20311 commented on Apr 30, 2025 • 0 new comments

• enqueue.cc:1556 NCCL WARN Cuda failure 700 'an illegal memory access was encountered'

  #20757 commented on May 1, 2025 • 0 new comments

• RTX5090d:ImportError: cannot import name 'EPOCH_OUTPUT' from 'pytorch_lightning.utilities.types'

  #20744 commented on May 2, 2025 • 0 new comments

• calling iter twice messes up dataloaders with queues

  #19427 commented on May 2, 2025 • 0 new comments

• Support gradient clipping by norm with FSDP

  #19235 commented on May 3, 2025 • 0 new comments