Weekly Patch Release v1.3.5 [full merge, no squash]

CHANGELOG.md

@@ -5,13 +5,33 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 
+## [1.3.5] - 2021-06-08
+
+### Added
+
+- Added warning to Training Step output ([#7779](https://github.com/PyTorchLightning/pytorch-lightning/pull/7779))
+
+### Fixed
+
+- Fixed LearningRateMonitor + BackboneFinetuning ([#7835](https://github.com/PyTorchLightning/pytorch-lightning/pull/7835))
+- Minor improvements to `apply_to_collection` and type signature of `log_dict` ([#7851](https://github.com/PyTorchLightning/pytorch-lightning/pull/7851))
+- Fixed docker versions ([#7834](https://github.com/PyTorchLightning/pytorch-lightning/pull/7834))
+- Fixed sharded training check for fp16 precision ([#7825](https://github.com/PyTorchLightning/pytorch-lightning/pull/7825))
+- Fixed support for torch Module type hints in LightningCLI ([#7807](https://github.com/PyTorchLightning/pytorch-lightning/pull/7807))
+
+### Changed
+
+- Move `training_output` validation to after `train_step_end` ([#7868](https://github.com/PyTorchLightning/pytorch-lightning/pull/7868))
+
+
 ## [1.3.4] - 2021-06-01
 
 ### Fixed
 
 - Fixed info message when max training time reached ([#7780](https://github.com/PyTorchLightning/pytorch-lightning/pull/7780))
 - Fixed missing `__len__` method to `IndexBatchSamplerWrapper` ([#7681](https://github.com/PyTorchLightning/pytorch-lightning/pull/7681))
 
+
 ## [1.3.3] - 2021-05-27
 
 ### Changed
@@ -41,8 +61,8 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Fixed setting correct `DistribType` for `ddp_cpu` (spawn) backend ([#7492](https://github.com/PyTorchLightning/pytorch-lightning/pull/7492))
 - Fixed incorrect number of calls to LR scheduler when `check_val_every_n_epoch > 1` ([#7032](https://github.com/PyTorchLightning/pytorch-lightning/pull/7032))
 
-## [1.3.1] - 2021-05-11
 
+## [1.3.1] - 2021-05-11
 
 ### Fixed
 
@@ -55,11 +75,11 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 ### Added
 
-- Added support for the `EarlyStopping` callback to run at the end of the training epoch ([#6944](https://github.com/PyTorchLightning/pytorch-lightning/pull/6944/))
+- Added support for the `EarlyStopping` callback to run at the end of the training epoch ([#6944](https://github.com/PyTorchLightning/pytorch-lightning/pull/6944))
 - Added synchronization points before and after `setup` hooks are run ([#7202](https://github.com/PyTorchLightning/pytorch-lightning/pull/7202))
 - Added a `teardown` hook to `ClusterEnvironment` ([#6942](https://github.com/PyTorchLightning/pytorch-lightning/pull/6942))
 - Added utils for metrics to scalar conversions ([#7180](https://github.com/PyTorchLightning/pytorch-lightning/pull/7180))
-- Added utils for NaN/Inf detection for gradients and parameters ([#6834](https://github.com/PyTorchLightning/pytorch-lightning/pull/6834/))
+- Added utils for NaN/Inf detection for gradients and parameters ([#6834](https://github.com/PyTorchLightning/pytorch-lightning/pull/6834))
 - Added more explicit exception message when trying to execute `trainer.test()` or `trainer.validate()` with `fast_dev_run=True` ([#6667](https://github.com/PyTorchLightning/pytorch-lightning/pull/6667))
 - Added `LightningCLI` class to provide simple reproducibility with minimum boilerplate training CLI (
     [#4492](https://github.com/PyTorchLightning/pytorch-lightning/pull/4492),
@@ -85,7 +105,7 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Added `outputs` parameter to callback's `on_validation_epoch_end` & `on_test_epoch_end` hooks ([#6120](https://github.com/PyTorchLightning/pytorch-lightning/pull/6120))
 - Added `configure_sharded_model` hook ([#6679](https://github.com/PyTorchLightning/pytorch-lightning/pull/6679))
 - Added support for `precision=64`, enabling training with double precision ([#6595](https://github.com/PyTorchLightning/pytorch-lightning/pull/6595))
-- Added support for DDP communication hooks ([#6736](https://github.com/PyTorchLightning/pytorch-lightning/issues/6736))
+- Added support for DDP communication hooks ([#6736](https://github.com/PyTorchLightning/pytorch-lightning/pull/6736))
 - Added `artifact_location` argument to `MLFlowLogger` which will be passed to the `MlflowClient.create_experiment` call ([#6677](https://github.com/PyTorchLightning/pytorch-lightning/pull/6677))
 - Added `model` parameter to precision plugins' `clip_gradients` signature (
     [#6764](https://github.com/PyTorchLightning/pytorch-lightning/pull/6764),
@@ -104,14 +124,14 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Added `tpu_distributed` check for TPU Spawn barrier ([#7241](https://github.com/PyTorchLightning/pytorch-lightning/pull/7241))
 - Added device updates to TPU Spawn for Pod training ([#7243](https://github.com/PyTorchLightning/pytorch-lightning/pull/7243))
 - Added warning when missing `Callback` and using `resume_from_checkpoint` ([#7254](https://github.com/PyTorchLightning/pytorch-lightning/pull/7254))
-- DeepSpeed single file saving ([#6900](https://github.com/PyTorchLightning/pytorch-lightning/issues/6900))
+- DeepSpeed single file saving ([#6900](https://github.com/PyTorchLightning/pytorch-lightning/pull/6900))
 - Added Training type Plugins Registry (
-    [#6982](https://github.com/PyTorchLightning/pytorch-lightning/issues/6982),
-    [#7063](https://github.com/PyTorchLightning/pytorch-lightning/issues/7063),
-    [#7214](https://github.com/PyTorchLightning/pytorch-lightning/issues/7214),
-    [#7224](https://github.com/PyTorchLightning/pytorch-lightning/issues/7224)
+    [#6982](https://github.com/PyTorchLightning/pytorch-lightning/pull/6982),
+    [#7063](https://github.com/PyTorchLightning/pytorch-lightning/pull/7063),
+    [#7214](https://github.com/PyTorchLightning/pytorch-lightning/pull/7214),
+    [#7224](https://github.com/PyTorchLightning/pytorch-lightning/pull/7224)
 )
-- Add `ignore` param to `save_hyperparameters` ([#6056](https://github.com/PyTorchLightning/pytorch-lightning/issues/6056))
+- Add `ignore` param to `save_hyperparameters` ([#6056](https://github.com/PyTorchLightning/pytorch-lightning/pull/6056))
 
 ### Changed
 
@@ -129,7 +149,7 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - The trainer no longer tries to save a checkpoint on exception or run callback's `on_train_end` functions ([#6864](https://github.com/PyTorchLightning/pytorch-lightning/pull/6864))
 - Changed `PyTorchProfiler` to use `torch.autograd.profiler.record_function` to record functions ([#6349](https://github.com/PyTorchLightning/pytorch-lightning/pull/6349))
 - Disabled `lr_scheduler.step()` in manual optimization  ([#6825](https://github.com/PyTorchLightning/pytorch-lightning/pull/6825))
-- Changed warnings and recommendations for dataloaders in `ddp_spawn` ([#6762](https://github.com/PyTorchLightning/pytorch-lightning/pull/6762/))
+- Changed warnings and recommendations for dataloaders in `ddp_spawn` ([#6762](https://github.com/PyTorchLightning/pytorch-lightning/pull/6762))
 - `pl.seed_everything` will now also set the seed on the `DistributedSampler` ([#7024](https://github.com/PyTorchLightning/pytorch-lightning/pull/7024))
 - Changed default setting for communication of multi-node training using `DDPShardedPlugin` ([#6937](https://github.com/PyTorchLightning/pytorch-lightning/pull/6937))
 - `trainer.tune()` now returns the tuning result ([#7258](https://github.com/PyTorchLightning/pytorch-lightning/pull/7258))
@@ -164,7 +184,7 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Deprecated the `save_function` property from the `ModelCheckpoint` callback ([#7201](https://github.com/PyTorchLightning/pytorch-lightning/pull/7201))
 - Deprecated `LightningModule.write_predictions` and `LightningModule.write_predictions_dict` ([#7066](https://github.com/PyTorchLightning/pytorch-lightning/pull/7066))
 - Deprecated `TrainerLoggingMixin` in favor of a separate utilities module for metric handling ([#7180](https://github.com/PyTorchLightning/pytorch-lightning/pull/7180))
-- Deprecated `TrainerTrainingTricksMixin` in favor of a separate utilities module for NaN/Inf detection for gradients and parameters ([#6834](https://github.com/PyTorchLightning/pytorch-lightning/pull/6834/))
+- Deprecated `TrainerTrainingTricksMixin` in favor of a separate utilities module for NaN/Inf detection for gradients and parameters ([#6834](https://github.com/PyTorchLightning/pytorch-lightning/pull/6834))
 - `period` has been deprecated in favor of `every_n_val_epochs` in the `ModelCheckpoint` callback ([#6146](https://github.com/PyTorchLightning/pytorch-lightning/pull/6146))
 - Deprecated `trainer.running_sanity_check` in favor of `trainer.sanity_checking` ([#4945](https://github.com/PyTorchLightning/pytorch-lightning/pull/4945))
 - Deprecated `Profiler(output_filename)` in favor of `dirpath` and `filename` ([#6621](https://github.com/PyTorchLightning/pytorch-lightning/pull/6621))
@@ -270,7 +290,7 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Fixed auto-scaling mode when calling tune method on trainer ([#7321](https://github.com/PyTorchLightning/pytorch-lightning/pull/7321))
 - Fixed finetuning complex models correctly unfreezes ([#6880](https://github.com/PyTorchLightning/pytorch-lightning/pull/6880))
 - Ensure we set the eval/train flag correctly on accelerator model ([#6877](https://github.com/PyTorchLightning/pytorch-lightning/pull/6877))
-- Set better defaults for `rank_zero_only.rank` when training is launched with SLURM and torchelastic ([#6802](https://github.com/PyTorchLightning/pytorch-lightning/pull/6802/))
+- Set better defaults for `rank_zero_only.rank` when training is launched with SLURM and torchelastic ([#6802](https://github.com/PyTorchLightning/pytorch-lightning/pull/6802))
 - Fixed matching the number of outputs of backward with forward for AllGatherGrad ([#6625](https://github.com/PyTorchLightning/pytorch-lightning/pull/6625))
 - Fixed the `gradient_clip_algorithm` has no effect ([#6928](https://github.com/PyTorchLightning/pytorch-lightning/pull/6928))
 - Fixed CUDA OOM detection and handling ([#6934](https://github.com/PyTorchLightning/pytorch-lightning/pull/6934))
@@ -486,7 +506,7 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Changed `computer_vision_fine_tunning` example to use `BackboneLambdaFinetuningCallback` ([#5377](https://github.com/PyTorchLightning/pytorch-lightning/pull/5377))
 - Changed `automatic casting` for LoggerConnector `metrics` ([#5218](https://github.com/PyTorchLightning/pytorch-lightning/pull/5218))
 - Changed `iou` [func] to allow float input ([#4704](https://github.com/PyTorchLightning/pytorch-lightning/pull/4704))
-- Metric `compute()` method will no longer automatically call `reset()` ([#5409](https://github.com/PyTorchLightning/pytorch-lightning/pull/5409/))
+- Metric `compute()` method will no longer automatically call `reset()` ([#5409](https://github.com/PyTorchLightning/pytorch-lightning/pull/5409))
 - Set PyTorch 1.4 as min requirements, also for testing and examples `torchvision>=0.5` and `torchtext>=0.5` ([#5418](https://github.com/PyTorchLightning/pytorch-lightning/pull/5418))
 - Changed `callbacks` argument in `Trainer` to allow `Callback` input ([#5446](https://github.com/PyTorchLightning/pytorch-lightning/pull/5446))
 - Changed the default of `find_unused_parameters` to `False` in DDP ([#5185](https://github.com/PyTorchLightning/pytorch-lightning/pull/5185))
@@ -1253,7 +1273,7 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Fixed getting `experiment_id` from MLFlow only once instead of each training loop ([#3394](https://github.com/PyTorchLightning/pytorch-lightning/pull/3394))
 - Fixed `overfit_batches` which now correctly disables shuffling for the training loader. ([#3501](https://github.com/PyTorchLightning/pytorch-lightning/pull/3501))
 - Fixed gradient norm tracking for `row_log_interval > 1` ([#3489](https://github.com/PyTorchLightning/pytorch-lightning/pull/3489))
-- Fixed `ModelCheckpoint` name formatting ([3164](https://github.com/PyTorchLightning/pytorch-lightning/pull/3163))
+- Fixed `ModelCheckpoint` name formatting ([#3164](https://github.com/PyTorchLightning/pytorch-lightning/pull/3163))
 - Fixed example implementation of AutoEncoder ([#3190](https://github.com/PyTorchLightning/pytorch-lightning/pull/3190))
 - Fixed invalid paths when remote logging with TensorBoard ([#3236](https://github.com/PyTorchLightning/pytorch-lightning/pull/3236))
 - Fixed change `t()` to `transpose()` as XLA devices do not support `.t()` on 1-dim tensor ([#3252](https://github.com/PyTorchLightning/pytorch-lightning/pull/3252))
@@ -1513,8 +1533,8 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Added option `save_last` to save the model at the end of every epoch in `ModelCheckpoint` ([#1908](https://github.com/PyTorchLightning/pytorch-lightning/pull/1908))
 - Early stopping checks `on_validation_end` ([#1458](https://github.com/PyTorchLightning/pytorch-lightning/pull/1458))
 - Speed up single-core TPU training by loading data using `ParallelLoader` ([#2033](https://github.com/PyTorchLightning/pytorch-lightning/pull/2033))
-- Added a model hook `transfer_batch_to_device` that enables moving custom data structures to the target device ([1756](https://github.com/PyTorchLightning/pytorch-lightning/pull/1756))
-- Added [black](https://black.readthedocs.io/en/stable/) formatter for the code with code-checker on pull ([1610](https://github.com/PyTorchLightning/pytorch-lightning/pull/1610))
+- Added a model hook `transfer_batch_to_device` that enables moving custom data structures to the target device ([#1756](https://github.com/PyTorchLightning/pytorch-lightning/pull/1756))
+- Added [black](https://black.readthedocs.io/en/stable/) formatter for the code with code-checker on pull ([#1610](https://github.com/PyTorchLightning/pytorch-lightning/pull/1610))
 - Added back the slow spawn ddp implementation as `ddp_spawn` ([#2115](https://github.com/PyTorchLightning/pytorch-lightning/pull/2115))
 - Added loading checkpoints from URLs ([#1667](https://github.com/PyTorchLightning/pytorch-lightning/pull/1667))
 - Added a callback method `on_keyboard_interrupt` for handling KeyboardInterrupt events during training ([#2134](https://github.com/PyTorchLightning/pytorch-lightning/pull/2134))

dockers/nvidia/Dockerfile

@@ -13,7 +13,7 @@
 # limitations under the License.
 
 # https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes
-FROM nvcr.io/nvidia/pytorch:21.04-py3
+FROM nvcr.io/nvidia/pytorch:21.05-py3
 
 LABEL maintainer="PyTorchLightning <https://github.com/PyTorchLightning>"
 
@@ -39,14 +39,16 @@ RUN \
 
 # Installations
     python -c "fname = './pytorch-lightning/requirements/extra.txt' ; lines = [line for line in open(fname).readlines() if not line.startswith('horovod')] ; open(fname, 'w').writelines(lines)" && \
+    pip install "Pillow>=8.2" "cryptography>=3.4" "py>=1.10" --no-cache-dir --upgrade-strategy only-if-needed && \
     pip install -r ./pytorch-lightning/requirements/extra.txt --no-cache-dir --upgrade-strategy only-if-needed && \
     pip install -r ./pytorch-lightning/requirements/examples.txt --no-cache-dir --upgrade-strategy only-if-needed && \
     pip install ./pytorch-lightning --no-cache-dir && \
-    pip install "Pillow>=8.1" --no-cache-dir --upgrade-strategy only-if-needed && \
     rm -rf pytorch-lightning && \
+    pip install jupyterlab[all] -U && \
     pip list
 
-RUN pip install lightning-grid -U
+RUN pip install lightning-grid -U && \
+    pip install "py>=1.10" "protobuf>=3.15.6" --upgrade-strategy only-if-needed
 
 ENV PYTHONPATH="/workspace"
 

pytorch_lightning/__about__.py

@@ -1,7 +1,7 @@
 import time
 
 _this_year = time.strftime("%Y")
-__version__ = '1.3.4'
+__version__ = '1.3.5'
 __author__ = 'William Falcon et al.'
 __author_email__ = 'waf2107@columbia.edu'
 __license__ = 'Apache-2.0'