Weekly Patch Release v1.5.10

[rohitgr7]

• Update version
• Fix apply_to_collection(defaultdict) (Fix apply_to_collection(defaultdict) #10316)
• Fix failure when DataLoader(batch_size=None) is passed (Fix failure when DataLoader(batch_size=None) is passed #10345)
• Update Python testing (Update Python testing #10269)
• Fix DataLoader inspection and re-instantiation in Lite (Fix DataLoader inspection and re-instantiation in Lite #10334)
• Only import PostLocalSGD related modules when it's needed (Only import PostLocalSGD related modules when it's needed #10359)
• Fix pickling error with CSVLogger (Fix pickling error with CSVLogger #10388)
• Revert part of Add custom dataloader support with Lite  #10279 (Revert part of #10279 #10376)
• disable step logging in epoch hooks (disable step logging in epoch hooks #10409)
• Fix deadlocks for distributed training for RichProgressBar (Fix deadlocks for distributed training for RichProgressBar #10428)
• Fix converting only float type tensors in Lite (Fix converting only float type tensors in Lite #10429)
• [Fault Tolerance] Don't check the len of a dataset, but its instance. ([Fault Tolerance] Don't check the len of a dataset, but its instance. #10432)
• Resolve workers being forcelly deleted with persistent_workers=True (Resolve workers being forcelly deleted with persistent_workers=True #10434)
• Update csv_logs.py
• [pre-commit.ci] auto fixes from pre-commit.com hooks
• update version
• Enable distributed training with CombinedDataLoader and max_size_cycle (Enable distributed training with CombinedDataLoader and max_size_cycle #10374)
• Fix support for dataclasses with ClassVar/InitVar in apply_to_collection (Fix support for dataclasses with ClassVar/InitVar in apply_to_collection #9702)
• fix overfit_batch sampler replacement logic (fix overfit_batch sampler replacement logic #10486)
• shutdown workers on failure (shutdown workers on failure #10463)
• Squeeze the early stopping monitor (Squeeze the early stopping monitor #10461)
• Resolve instantiation problem with init_meta_context (Resolve instantiation problem with init_meta_context #10493)
• Fix to_torchscript() causing false positive deprecation warnings (Fix to_torchscript() causing false positive deprecation warnings #10470)
• Skip strategy=ddp_spawn, accelerator=cpu, python>=3.9 tests (Skip strategy=ddp_spawn, accelerator=cpu, python>=3.9 tests #10550)
• Fix scripting causing false positive deprecation warnings (Fix scripting causing false positive deprecation warnings #10555)
• Change attributes of RichProgressBarTheme dataclass (Change attributes of RichProgressBarTheme dataclass #10454)
• 1.5.2 release
• [DeepSpeed] Do not fail if batch size could not be inferred for logging ([DeepSpeed] Do not fail if batch size could not be inferred for logging #10438)
• Fix propagation of device and dtype properties in Lite modules (Fix propagation of device and dtype properties in Lite modules #10559)
• update version
• Remove redundant fit call from accelerator connector test (Remove redundant fit call from accelerator connector test #10626)
• Check torch.distributed availability before sharded tensor state dict hook registration (Check torch.distributed availability before sharded tensor state dict hook registration #10621)
• Lite: Don't pop value if they don't exist (Lite: Don't pop value if they don't exist #10613)
• Fix batch size extraction when set by the user in LightningModule.log (Fix an issue that caused Lightning to extract the batch size even though it was set by the user in LightningModule.log #10408)
• Respect the passed dtype with self.log (Respect the passed dtype with self.log #10076)
• Simplify hanging queue test (Simplify hanging queue test #10591)
• Added boring model as a ipynb so it can be updated (Added boring model as a ipynb so it can be updated #10521)
• Fix misleading ModelCheckpoint documentation on every_n_epochs parameter (Fix documentation for every_n_epochs parameter of ModelCheckpoint #10421)
• update bug_report model links and notebook (update bug_report model links and notebook #10665)
• Remove metrics references from docs (Remove metrics references from docs #10567)
• Fix docs filterwarnings snippet (Fix docs filterwarnings snippet #10671)
• log metrics for correct dataloader only (log metrics for correct dataloader only #10522)
• Fix move_metrics_to_cpu with evaluation (Fix move_metrics_to_cpu with evaluation #10631)
• Don't register signal in thread (Don't register signal in thread #10610)
• MANIFEST.in and setup.py clean-up (MANIFEST.in and setup.py clean-up #7614)
• Support special test parametrizations (Support special test parametrizations #10569)
• Move benchmarks into the test directory (Move benchmarks into the test directory #10614)
• 1.5.3 release notes
• Merge pull request Sync CI configuration into 1.5.x branch #10738 from PyTorchLightning/1.5.x-drop-torch-1.6
• update version
• Fix compare version for packages (Fix compare version for packages #10762)
• Delete TensorBoardLogger experiment before spawning the processes. (Delete TensorBoardLogger experiment before spawning the processes. #10777)
• Consolidate state when retrieving sharded state dict in Lite (Consolidate state when retrieving sharded state dict in Lite #10746)
• Fix default logging levels for train step specific hooks (Fix default logging levels for train step specific hooks #10756)
• Update notebooks submodule (Update notebooks submodule #10827)
• [CLI] Add support for --key.help=class
• Tune Conda CI timeout and other minor improvements (Tune Conda CI timeout and other minor improvements #10769)
• Rename special to standalone (Rename special to standalone #10779)
• Merge pull request Sync CI configuration into 1.5.x branch #10738 from PyTorchLightning/1.5.x-drop-torch-1.6
• Update changelog
• update version
• Disable batch_size extraction for torchmetric instances (Disable batch_size extraction for torchmetric instances #10815)
• Fix selection of standalone tests (Fix selection of standalone tests #10857)
• Restore signals on teardown (Restore signals on teardown #10611)
• Fix SignalConnector._has_already_handler check for callable type (Fix SignalConnector._has_already_handler check for callable type #10483)
• Skip hanging spawn tests (Skip hanging spawn tests #10838)
• Raise exception if rich is less than 10.2.2 (Raise exception if rich is less than 10.2.2 #10839)
• Fixed uploading best model checkpoint in NeptuneLogger (Fixed uploading best model checkpoint in NeptuneLogger #10369)
• Fix schedule reset logic in pytorch profiler (Fix schedule reset logic in pytorch profiler #10837)
• Fix retrieval of batch indices when dataloader num_workers > 0 (Fix retrieval of batch indices when dataloader num_workers > 0 #10870)
• Fix filtration logic for eval results with multiple dataloaders (Fix filteration logic for eval results #10810)
• Simplify some ddp-spawn tests Simplify some ddp-spawn tests #10921
• Resolve: 'DummyExperiment' object does not support item assignment (Resolve: 'DummyExperiment' object does not support item assignment #10917)
• Don't import torch_xla.debug for torch-xla<1.8 (Don't import torch_xla.debug for torch-xla<1.8 #10836)
• triger ci only with pull request (Trigger TPU CI only on branches with a pull request #10932)
• Fix spawn plugins not deleting temp checkpoint (Fix spawn plugins not deleting temp checkpoint #10935)
• fix TypeError cause failure in singal_connector teardown (fix TypeError cause failure in singal_connector teardown #10961)
• 1.5.5 release
• update chlog
• [DeepSpeed] fix flag forwarding in DeepSpeedPlugin ([DeepSpeed] fix flag forwarding in DeepSpeedPlugin #10899)
• Removed duplicated file extension when uploading model checkpoints with NeptuneLogger (Removed duplicated file extension when uploading model checkpoints with NeptuneLogger #11015)
• Fix support for logging within callbacks returned from LightningModule (Fix support for logging within callbacks returned from LightningModule #10991)
• Fix the num_batches value in warning (Fix the num_batches value in error message #10980)
• Fix support for CombinedLoader while checking for warning raised with eval dataloaders (Fix support for CombinedLoader while checking for warning raised with eval dataloaders #10994)
• Fix sanity check for RichProgressBar (Fix sanity check for RichProgressBar #10913)
• Update the TQDM progress bar on_train_epoch_end (Update the TQDM progress bar on_train_epoch_end #11069)
• update version to 1.5.7
• Support torch 1.10.1 (Support torch 1.10.1 #11095)
• Update mypy (Update mypy to 0.920 #11096)
• Teardown sync-batchnorm after training (Teardown sync-batchnorm after training #11078)
• Fix signal teardown outside main thread (Fix signal teardown outside main thread #11124)
• Fix AttributeError when using CombinedLoader in prediction (Fix AttributeError when using CombinedLoader in prediction #11111)
• Enable logging hparams only if there are any (Enable logging hparams only if there are any #11105)
• Fix tpu spawn plugin test (Fix tpu spawn plugin test #11131)
• Fixed NeptuneLogger when using DDP (Fixed NeptuneLogger when using DDP #11030)
• Avoid the deprecated onnx.export(example_outputs=...) in torch 1.10 (Avoid the deprecated onnx.export(example_outputs=...) in torch 1.10 #11116)
• Remove partitioning of model in ZeRO 3 ([DeepSpeed] Remove partitioning of model in ZeRO 3 #10655)
• Add typing for trainer.logger (Add typing for trainer.logger #11114)
• Refactor plugin tests whose assertions don't need to run in on_fit_start hook (Refactor plugin tests whose assertions don't need to run in on_fit_start hook #11149)
• Safely disable profiler (Safely disable profiler #11167)
• Add required states for resumed ModelCheckpoint GC (Add required states for resumed ModelCheckpoint GC #10995)
• Avoid torch amp cuda warning with bf16 on cpu (Avoid torch amp cuda warning with bf16 on cpu #11161)
• Suppress Warning in PredictionEpochLoop (Suppress Warning in PredictionEpochLoop #11189)
• update version for 1.5.8 release
• Fix CLI race condition saving the config (Fix CLI race condition saving the config #11199)
• Revert "[CI] Comment flaky tests ([CI] Comment flaky tests #10084)" (Revert "[CI] Comment flaky tests (#10084)" #10580)
• Fix data fetcher selection (Fix data fetcher selection #11294)
• Avoid non-blocking GPU->CPU copies. (Avoid non-blocking GPU->CPU copies. #11288)
• 1.5.8 release commit
• Fix _should_reload_dl_epoch causing inconsistent validation dataloader reloading (Fix _should_reload_dl_epoch causing inconsistent validation dataloader reloading #11036)
• Fix min/max logging default value (Fix min/max logging default value #11310)
• Update Version
• pin sphinx-autodoc-typehints version to v1.15 (Pin sphinx-autodoc-typehints with <v1.15 #11400)
• Skip testing with PyTorch 1.7 and Python 3.9 on Ubuntu (Skip testing with PyTorch 1.7 and Python 3.9 on Ubuntu #11217)
• release 1.5.9
• Avoid in-place ops during logging result updates (Avoid in-place ops during logging result updates #11401)
• Modify LSFEnvironment to use more reliable environment variable (Modify LSFEnvironment to use more reliable environment variable #10825)
• Disable attaching samplers when using IterableDataset (Disable attaching samplers when using IterableDataset #11507)
• Pin jsonargparse maximum version
• [CLI] Save only the configuration used ([CLI] Save only the configuration used #11532)
• Don't use old testing packages in CI (Don't use old testing packages in CI #11366)
• Update CHANGELOG.md
• update version

What does this PR do?

Fixes #<issue_number>

Does your PR introduce any breaking changes? If yes, please list them.

Before submitting

• Was this discussed/approved via a GitHub issue? (not for typos and docs)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure your PR does only one thing, instead of bundling different changes together?
• Did you make sure to update the documentation with your changes? (if necessary)
• Did you write any new necessary tests? (not for typos and docs)
• Did you verify new and existing tests pass locally with your changes?
• Did you list all the breaking changes introduced by this pull request?
• Did you update the CHANGELOG? (not for typos, docs, test updates, or internal minor changes/refactorings)

PR review

Anyone in the community is welcome to review the PR.
Before you start reviewing make sure you have read Review guidelines. In short, see the following bullet-list:

• Is this pull request ready for review? (if not, please submit in draft mode)
• Check that all items from Before submitting are resolved
• Make sure the title is self-explanatory and the description concisely explains the PR
• Add labels and milestones (and optionally projects) to the PR so it can be classified

Did you have fun?

Make sure you had fun coding 🙃

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB