Lightning

Fixed

- Fixed the availability check for the neptune-client package ([14714](https://github.com/Lightning-AI/lightning/pull/14714))
- Break HPU Graphs into two parts (forward + backward as one and optimizer as another) for better performance ([14656](https://github.com/Lightning-AI/lightning/pull/14656))
- Fixed torchscript error with ensembles of LightningModules ([14657](https://github.com/Lightning-AI/lightning/pull/14657), [#14724](https://github.com/Lightning-AI/lightning/pull/14724))
- Fixed an issue with `TensorBoardLogger.finalize` creating a new experiment when none was created during the Trainer's execution ([14762](https://github.com/Lightning-AI/lightning/pull/14762))
- Fixed `TypeError` on import when `torch.distributed` is not available ([14809](https://github.com/Lightning-AI/lightning/pull/14809))

Contributors

awaelchli Borda carmocca dependabot otaj raoakarsha

_If we forgot someone due to not matching commit email with GitHub account, let us know_ :)

Changed

- Improved the error messaging when passing `Trainer.method(model, x_dataloader=None)` with no module-method implementations available ([14614](https://github.com/Lightning-AI/lightning/pull/14614))

Fixed

- Reset the dataloaders on OOM failure in batch size finder to use the last successful batch size ([14372](https://github.com/Lightning-AI/lightning/pull/14372))
- Fixed an issue to keep downscaling the batch size in case there hasn't been even a single successful optimal batch size with `mode="power"` ([14372](https://github.com/Lightning-AI/lightning/pull/14372))
- Fixed an issue where `self.log`-ing a tensor would create a user warning from PyTorch about cloning tensors ([14599](https://github.com/Lightning-AI/lightning/pull/14599))
- Fixed compatibility when `torch.distributed` is not available ([14454](https://github.com/Lightning-AI/lightning/pull/14454))

Contributors

akihironitta awaelchli Borda carmocca dependabot krshrimali mauvilsa pierocor rohitgr7 wangraying

_If we forgot someone due to not matching commit email with GitHub account, let us know_ :)

Fixed

- Squeezed tensor values when logging with `LightningModule.log` ([14489](https://github.com/Lightning-AI/lightning/pull/14489))
- Fixed `WandbLogger` `save_dir` is not set after creation ([14326](https://github.com/Lightning-AI/lightning/pull/14326))
- Fixed `Trainer.estimated_stepping_batches` when maximum number of epochs is not set ([14317](https://github.com/Lightning-AI/lightning/pull/14317))

Contributors

carmocca dependabot robertomest rohitgr7 tshu-w

_If we forgot someone due to not matching commit email with GitHub account, let us know_ :)

Added

- Added an environment variable `PL_DISABLE_FORK` that can be used to disable all forking in the Trainer ([14319](https://github.com/Lightning-AI/lightning/issues/14319))

Fixed

- Fixed `LightningDataModule` hparams parsing ([12806](https://github.com/PyTorchLightning/pytorch-lightning/pull/12806))
- Reset epoch progress with batch size scaler ([13846](https://github.com/Lightning-AI/lightning/pull/13846))
- Fixed restoring the trainer after using `lr_find()` so that the correct LR schedule is used for the actual training ([14113](https://github.com/Lightning-AI/lightning/pull/14113))
- Fixed incorrect values after transferring data to an MPS device ([14368](https://github.com/Lightning-AI/lightning/pull/14368))

Contributors

rohitgr7 tanmoyio justusschock cschell carmocca Callidior awaelchli j0rd1smit dependabot Borda otaj

Fixed

- Fixed an assertion error when using a `ReduceOnPlateau` scheduler with the Horovod strategy ([14215](https://github.com/Lightning-AI/lightning/pull/14215))
- Fixed an `AttributeError` when accessing `LightningModule.logger` and the Trainer has multiple loggers ([14234](https://github.com/Lightning-AI/lightning/pull/14234))
- Fixed wrong num padding for `RichProgressBar` ([14296](https://github.com/Lightning-AI/lightning/pull/14296))
- Added back support for `log`ging in the `configure_gradient_clipping` hook after unintended removal in v1.7.2 ([14298](https://github.com/Lightning-AI/lightning/issues/14298))
- Fixed an issue to avoid the impact of sanity check on `reload_dataloaders_every_n_epochs` for validation ([13964](https://github.com/Lightning-AI/lightning/pull/13964))


Contributors

awaelchli Borda carmocca dependabot kaushikb11 otaj rohitgr7

Added

- Added `FullyShardedNativeNativeMixedPrecisionPlugin` to handle precision for `DDPFullyShardedNativeStrategy` ([14092](https://github.com/Lightning-AI/lightning/pull/14092))
- Added profiling to these hooks: `on_before_batch_transfer`, `transfer_batch_to_device`, `on_after_batch_transfer`, `configure_gradient_clipping`, `clip_gradients` ([14069](https://github.com/Lightning-AI/lightning/pull/14069))

Changed

- Updated compatibility for LightningLite to run with the latest DeepSpeed 0.7.0 ([13967](https://github.com/Lightning-AI/lightning/pull/13967))
- Raised a `MisconfigurationException` if batch transfer hooks are overriden with `IPUAccelerator` ([13961](https://github.com/Lightning-AI/lightning/pull/13961))
- The default project name in `WandbLogger` is now "lightning_logs" ([14145](https://github.com/Lightning-AI/lightning/pull/14145))
- The `WandbLogger.name` property no longer returns the name of the experiment, and instead returns the project's name ([14145](https://github.com/Lightning-AI/lightning/pull/14145))

Fixed

- Fixed a bug that caused spurious `AttributeError` when multiple `DataLoader` classes are imported ([14117](https://github.com/Lightning-AI/lightning/pull/14117))
- Fixed epoch-end logging results not being reset after the end of the epoch ([14061](https://github.com/Lightning-AI/lightning/pull/14061))
- Fixed saving hyperparameters in a composition where the parent class is not a `LightningModule` or `LightningDataModule` ([14151](https://github.com/Lightning-AI/lightning/pull/14151))
- Fixed epoch-end logging results not being reset after the end of the epoch ([14061](https://github.com/Lightning-AI/lightning/pull/14061))
- Fixed the device placement when `LightningModule.cuda()` gets called without specifying a device index and the current cuda device was not 0 ([14128](https://github.com/Lightning-AI/lightning/pull/14128))
- Avoided false positive warning about using `sync_dist` when using torchmetrics ([14143](https://github.com/Lightning-AI/lightning/pull/14143))
- Avoid `metadata.entry_points` deprecation warning on Python 3.10 ([14052](https://github.com/Lightning-AI/lightning/pull/14052))
- Avoid raising the sampler warning if num_replicas=1 ([14097](https://github.com/Lightning-AI/lightning/pull/14097))
- Fixed resuming from a checkpoint when using Stochastic Weight Averaging (SWA) ([9938](https://github.com/Lightning-AI/lightning/pull/9938))
- Avoided requiring the FairScale package to use precision with the fsdp native strategy ([14092](https://github.com/Lightning-AI/lightning/pull/14092))
- Fixed an issue in which the default name for a run in `WandbLogger` would be set to the project name instead of a randomly generated string ([14145](https://github.com/Lightning-AI/lightning/pull/14145))
- Fixed not preserving set attributes on `DataLoader` and `BatchSampler` when instantiated inside `*_dataloader` hooks ([14212](https://github.com/Lightning-AI/lightning/pull/14212))


Contributors

adamreeve akihironitta awaelchli Borda carmocca dependabot otaj rohitgr7