[release] Weekly Patch Release v.1.2.3 [full merge, no squash]

.github/workflows/ci_test-conda.yml

@@ -27,24 +27,20 @@ jobs:
       run: |
         conda info
         conda list
+        # adjust versions according installed Torch version
+        python ./requirements/adjust_versions.py requirements/extra.txt
+        python ./requirements/adjust_versions.py requirements/examples.txt
         pip install --requirement requirements/devel.txt --upgrade-strategy only-if-needed
         pip list
 
     - name: Pull checkpoints from S3
-      # todo: consider adding coma caching, but ATM all models have less then 100KB
       run: |
         # enter legacy and update checkpoints from S3
         cd legacy
         curl https://pl-public-data.s3.amazonaws.com/legacy/checkpoints.zip --output checkpoints.zip
         unzip -o checkpoints.zip
         ls -l checkpoints/
 
-    # todo: require proper fix in docker image
-    - name: Hotfix dependency
-      run: |
-        pip install torchtext==0.6.0 -U
-      shell: bash
-
     - name: Tests
       run: |
         # NOTE: run coverage on tests does not propagare faler status for Win, https://github.com/nedbat/coveragepy/issues/1003

.github/workflows/ci_test-full.yml

@@ -104,20 +104,17 @@ jobs:
         HOROVOD_WITHOUT_MXNET: 1
         HOROVOD_WITHOUT_TENSORFLOW: 1
       run: |
-        # python -m pip install --upgrade --user pip
-        pip install --requirement requirements.txt --find-links https://download.pytorch.org/whl/cpu/torch_stable.html --quiet --upgrade
-        pip install --requirement ./requirements/devel.txt --find-links https://download.pytorch.org/whl/cpu/torch_stable.html --quiet --upgrade
         python --version
         pip --version
+        # python -m pip install --upgrade --user pip
+        pip install --requirement requirements.txt --find-links https://download.pytorch.org/whl/cpu/torch_stable.html --upgrade
+        # adjust versions according installed Torch version
+        python ./requirements/adjust_versions.py requirements/extra.txt
+        python ./requirements/adjust_versions.py requirements/examples.txt
+        pip install --requirement ./requirements/devel.txt --find-links https://download.pytorch.org/whl/cpu/torch_stable.html --upgrade
         pip list
       shell: bash
 
-    # todo: require proper fix in docker image
-    - name: Hotfix dependency
-      run: |
-        pip install torchtext==0.6.0 -U
-      shell: bash
-
     - name: Reinstall Horovod if necessary
       if: runner.os != 'windows'
       env:
@@ -143,10 +140,9 @@ jobs:
         # NOTE: do not include coverage report here, see: https://github.com/nedbat/coveragepy/issues/1003
         coverage run --source pytorch_lightning -m pytest pytorch_lightning tests -v --durations=50 --junitxml=junit/test-results-${{ runner.os }}-${{ matrix.python-version }}-${{ matrix.requires }}.xml
 
-    # todo: put this back just when TorchVision can download datasets
-    #- name: Examples
-    #  run: |
-    #    python -m pytest pl_examples -v --durations=10
+    - name: Examples
+      run: |
+        python -m pytest pl_examples -v --durations=10
 
     - name: Upload pytest test results
       uses: actions/upload-artifact@v2

.github/workflows/docs-checks.yml

@@ -41,15 +41,15 @@ jobs:
 
       - name: Install dependencies
         run: |
+          python --version
+          pip --version
           # remove Horovod from requirements
           python -c "fname = 'requirements/extra.txt' ; lines = [line for line in open(fname).readlines() if not line.startswith('horovod')] ; open(fname, 'w').writelines(lines)"
           # python -m pip install --upgrade --user pip
           pip install --requirement requirements.txt --upgrade-strategy only-if-needed --find-links https://download.pytorch.org/whl/cpu/torch_stable.html --quiet
           pip install --requirement requirements/extra.txt
           pip install --requirement requirements/loggers.txt
           pip install --requirement requirements/docs.txt
-          python --version
-          pip --version
           pip list
         shell: bash
 
@@ -84,12 +84,12 @@ jobs:
 
       - name: Install dependencies
         run: |
-          pip install --requirement requirements.txt --upgrade-strategy only-if-needed --find-links https://download.pytorch.org/whl/cpu/torch_stable.html --quiet
+          python --version
+          pip --version
+          # pip install --requirement requirements.txt --upgrade-strategy only-if-needed --find-links https://download.pytorch.org/whl/cpu/torch_stable.html --quiet
           pip install --requirement requirements/docs.txt
           # install Texlive, see https://linuxconfig.org/how-to-install-latex-on-ubuntu-20-04-focal-fossa-linux
           sudo apt-get update && sudo apt-get install -y texlive-latex-extra dvipng texlive-pictures
-          python --version
-          pip --version
           pip list
         shell: bash
 

.github/workflows/events-nightly.yml

@@ -102,8 +102,6 @@ jobs:
         id: extend
 
       - name: Publish CUDA to Docker Hub
-        # ToDo: extend also building for Nightly from pip
-        if: matrix.pytorch_version < 1.8
         # publish master/release
         uses: docker/build-push-action@v2
         with:

CHANGELOG.md

@@ -5,6 +5,24 @@ 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/).
 
 
+## [1.2.3] - 2021-03-09
+
+
+### Fixed
+
+- Fixed `ModelPruning(make_pruning_permanent=True)` pruning buffers getting removed when saved during training ([#6073](https://github.com/PyTorchLightning/pytorch-lightning/pull/6073))
+- Fixed when `_stable_1d_sort` to work when `n >= N` ([#6177](https://github.com/PyTorchLightning/pytorch-lightning/pull/6177))
+- Fixed `AttributeError` when `logger=None` on TPU ([#6221](https://github.com/PyTorchLightning/pytorch-lightning/pull/6221))
+- Fixed PyTorch Profiler with `emit_nvtx` ([#6260](https://github.com/PyTorchLightning/pytorch-lightning/pull/6260))
+- Fixed `trainer.test` from `best_path` hangs after calling `trainer.fit`  ([#6272](https://github.com/PyTorchLightning/pytorch-lightning/pull/6272))
+- Fixed `SingleTPU` calling `all_gather` ([#6296](https://github.com/PyTorchLightning/pytorch-lightning/pull/6296))
+- Ensure we check deepspeed/sharded in multinode DDP ([#6297](https://github.com/PyTorchLightning/pytorch-lightning/pull/6297)
+- Check `LightningOptimizer` doesn't delete optimizer hooks ([#6305](https://github.com/PyTorchLightning/pytorch-lightning/pull/6305)
+- Resolve memory leak for evaluation ([#6326](https://github.com/PyTorchLightning/pytorch-lightning/pull/6326)
+- Ensure that clip gradients is only called if the value is greater than 0 ([#6330](https://github.com/PyTorchLightning/pytorch-lightning/pull/6330)
+- Fixed `Trainer` not resetting `lightning_optimizers` when calling `Trainer.fit()` multiple times ([#6372](https://github.com/PyTorchLightning/pytorch-lightning/pull/6372))
+
+
 ## [1.2.2] - 2021-03-02
 
 ### Added

MANIFEST.in

@@ -46,7 +46,7 @@ recursive-include docs/source/_static/images/general/ pl_overview* tf_* tutorial
 
 # Include the Requirements
 recursive-include requirements *.txt
-recursive-exclude requirements *.sh
+recursive-exclude requirements *.sh *.py
 include requirements.txt
 include pyproject.toml
 

azure-pipelines.yml

@@ -71,11 +71,6 @@ jobs:
         python -c "import torch ; mgpu = torch.cuda.device_count() ; assert mgpu >= 2, f'GPU: {mgpu}'"
       displayName: 'Env details'
 
-    # todo: require proper fix in docker image
-    - bash: |
-        pip install torchtext==0.7 -U
-      displayName: 'HotFix'
-
     - bash: |
         wget https://pl-public-data.s3.amazonaws.com/legacy/checkpoints.zip -P legacy/
         unzip -o legacy/checkpoints.zip -d legacy/
@@ -100,10 +95,12 @@ jobs:
         python -m pytest benchmarks -v --maxfail=2 --durations=0
       displayName: 'Testing: benchmarks'
 
-    # todo: put this back just when TorchVision can download datasets
-    #- bash: |
-    #    python -m pytest pl_examples -v --maxfail=2 --durations=0
-    #    python setup.py install --user --quiet
-    #    bash pl_examples/run_ddp-example.sh
-    #    pip uninstall -y pytorch-lightning
-    #  displayName: 'Examples'
+    - bash: |
+        python -m pytest pl_examples -v --maxfail=2 --durations=0
+        python setup.py install --user --quiet
+        bash pl_examples/run_ddp-example.sh
+        cd pl_examples/basic_examples
+        bash submit_ddp_job.sh
+        bash submit_ddp2_job.sh
+        pip uninstall -y pytorch-lightning
+      displayName: 'Examples'

dockers/base-conda/Dockerfile

@@ -98,10 +98,12 @@ ENV \
 
 COPY ./requirements/extra.txt requirements-extra.txt
 COPY ./requirements/test.txt requirements-test.txt
+COPY ./requirements/adjust_versions.py requirements_adjust_versions.py
 
 RUN \
     pip list | grep torch && \
     python -c "import torch; print(torch.__version__)" && \
+    python requirements_adjust_versions.py requirements-extra.txt && \
     # Install remaining requirements
     pip install -r requirements-extra.txt --no-cache-dir && \
     pip install -r requirements-test.txt --no-cache-dir && \

dockers/base-cuda/Dockerfile

@@ -94,12 +94,14 @@ RUN \
     # eventualy use pre-release
     #pip install "torch==${PYTORCH_VERSION}.*" --pre && \
     # set particular PyTorch version
-    python -c "import re ; fname = 'requirements.txt' ; req = re.sub(r'torch[>=]+[\d\.]+', 'torch==${PYTORCH_VERSION}.*', open(fname).read()) ; open(fname, 'w').write(req)" && \
+    python ./requirements/adjust_versions.py requirements.txt ${PYTORCH_VERSION} && \
+    python ./requirements/adjust_versions.py requirements/extra.txt ${PYTORCH_VERSION} && \
+    python ./requirements/adjust_versions.py requirements/examples.txt ${PYTORCH_VERSION} && \
     # Install all requirements
     # todo: find a way how to install nightly PT version
     #  --pre --extra-index-url https://download.pytorch.org/whl/nightly/cu${cuda_ver[0]}${cuda_ver[1]}/torch_nightly.html
     pip install -r requirements/devel.txt --no-cache-dir && \
-    rm -rf requirements*
+    rm -rf requirements.* requirements/
 
 RUN \
     # install DALI, needed for examples
@@ -113,7 +115,7 @@ RUN \
 
 RUN \
     # install DeepSpeed from source.
-    # todo: swap to pypi release once DeepSpeed releases a new version.
+    # todo: swap to pypi release once DeepSpeed releases a new version >= 0.3.10
     pip install deepspeed@git+https://github.com/microsoft/DeepSpeed@ec8b1cb
 
 RUN \

dockers/base-xla/Dockerfile

@@ -104,6 +104,7 @@ RUN \
     python -c "fname = 'requirements/extra.txt' ; lines = [line for line in open(fname).readlines() if 'fairscale' not in line] ; open(fname, 'w').writelines(lines)" && \
     # drop TorchVision as it was installed with XLA
     python -c "fname = 'requirements/examples.txt' ; lines = [line for line in open(fname).readlines() if not line.startswith('torchvision')] ; open(fname, 'w').writelines(lines)" && \
+    python ./requirements/adjust_versions.py ./requirements/extra.txt && \
     pip install --requirement ./requirements/devel.txt --no-cache-dir && \
     cd .. && \
     rm -rf pytorch-lightning && \

dockers/release/Dockerfile

@@ -27,14 +27,15 @@ COPY ./ ./pytorch-lightning/
 RUN \
     # Disable cache
     #conda install "pip>20.1" && \
-    #pip config set global.cache-dir false && \
-    if [ -z $LIGHTNING_VERSION ] ; then \
-        pip install ./pytorch-lightning --no-cache-dir ; \
+    if [ ! -z "$LIGHTNING_VERSION" ] ; then \
         rm -rf pytorch-lightning ; \
-    else \
-        rm -rf pytorch-lightning ; \
-        pip install https://github.com/PyTorchLightning/pytorch-lightning/archive/${LIGHTNING_VERSION}.zip --no-cache-dir ; \
-    fi
+        wget https://github.com/PyTorchLightning/pytorch-lightning/archive/${LIGHTNING_VERSION}.zip --progress=bar:force:noscroll ; \
+        unzip ${LIGHTNING_VERSION}.zip ; \
+        mv pytorch-lightning-*/ pytorch-lightning ; \
+        rm *.zip ; \
+    fi && \
+    pip install ./pytorch-lightning["extra"] --no-cache-dir && \
+    rm -rf pytorch-lightning
 
 RUN python --version && \
     pip --version && \

docs/source/advanced/multiple_loaders.rst

@@ -16,6 +16,8 @@ Lightning supports multiple dataloaders in a few ways.
 
 ----------
 
+.. _multiple-training-dataloaders:
+
 Multiple training dataloaders
 -----------------------------
 For training, the usual way to use multiple dataloaders is to create a ``DataLoader`` class
@@ -86,6 +88,27 @@ For more details please have a look at :attr:`~pytorch_lightning.trainer.trainer
 
             return loaders
 
+Furthermore, Lightning also supports that nested lists and dicts (or a combination) can
+be returned 
+
+.. testcode::
+
+    class LitModel(LightningModule):
+
+        def train_dataloader(self):
+
+            loader_a = torch.utils.data.DataLoader(range(8), batch_size=4)
+            loader_b = torch.utils.data.DataLoader(range(16), batch_size=4)
+            loader_c = torch.utils.data.DataLoader(range(32), batch_size=4)
+            loader_c = torch.utils.data.DataLoader(range(64), batch_size=4)
+
+            # pass loaders as a nested dict. This will create batches like this:
+            # {'loader_a_b': {'a': batch from loader a, 'b': batch from loader b},
+            #  'loader_c_d': {'c': batch from loader c, 'd': batch from loader d}}
+            loaders = {'loaders_a_b': {'a': loader_a, 'b': loader_b},
+                       'loaders_c_d': {'c': loader_c, 'd': loader_d}}
+            return loaders
+
 ----------
 
 Test/Val dataloaders

notebooks/06-mnist-tpu-training.ipynb

@@ -80,7 +80,7 @@
         "id": "AYGWh10lRaF1"
       },
       "source": [
-        "! pip install cloud-tpu-client==0.10 https://storage.googleapis.com/tpu-pytorch/wheels/torch_xla-1.7-cp36-cp36m-linux_x86_64.whl"
+        "! pip install cloud-tpu-client==0.10 https://storage.googleapis.com/tpu-pytorch/wheels/torch_xla-1.7-cp37-cp37m-linux_x86_64.whl"
       ],
       "execution_count": null,
       "outputs": []

pl_examples/__init__.py

@@ -1,14 +1,30 @@
 import os
+from urllib.error import HTTPError
+
+from six.moves import urllib
 
 from pytorch_lightning.utilities import _module_available
 
+# TorchVision hotfix https://github.com/pytorch/vision/issues/1938
+opener = urllib.request.build_opener()
+opener.addheaders = [('User-agent', 'Mozilla/5.0')]
+urllib.request.install_opener(opener)
+
 _EXAMPLES_ROOT = os.path.dirname(__file__)
 _PACKAGE_ROOT = os.path.dirname(_EXAMPLES_ROOT)
 _DATASETS_PATH = os.path.join(_PACKAGE_ROOT, 'Datasets')
 
 _TORCHVISION_AVAILABLE = _module_available("torchvision")
+_TORCHVISION_MNIST_AVAILABLE = True
 _DALI_AVAILABLE = _module_available("nvidia.dali")
 
+if _TORCHVISION_AVAILABLE:
+    try:
+        from torchvision.datasets.mnist import MNIST
+        MNIST(_DATASETS_PATH, download=True)
+    except HTTPError:
+        _TORCHVISION_MNIST_AVAILABLE = False
+
 LIGHTNING_LOGO = """
                     ####
                 ###########

pl_examples/basic_examples/autoencoder.py

@@ -20,9 +20,9 @@
 from torch.utils.data import DataLoader, random_split
 
 import pytorch_lightning as pl
-from pl_examples import _DATASETS_PATH, _TORCHVISION_AVAILABLE, cli_lightning_logo
+from pl_examples import _DATASETS_PATH, _TORCHVISION_AVAILABLE, _TORCHVISION_MNIST_AVAILABLE, cli_lightning_logo
 
-if _TORCHVISION_AVAILABLE:
+if _TORCHVISION_AVAILABLE and _TORCHVISION_MNIST_AVAILABLE:
     from torchvision import transforms
     from torchvision.datasets.mnist import MNIST
 else:

pl_examples/basic_examples/backbone_image_classifier.py

@@ -19,9 +19,9 @@
 from torch.utils.data import DataLoader, random_split
 
 import pytorch_lightning as pl
-from pl_examples import _DATASETS_PATH, _TORCHVISION_AVAILABLE, cli_lightning_logo
+from pl_examples import _DATASETS_PATH, _TORCHVISION_AVAILABLE, _TORCHVISION_MNIST_AVAILABLE, cli_lightning_logo
 
-if _TORCHVISION_AVAILABLE:
+if _TORCHVISION_AVAILABLE and _TORCHVISION_MNIST_AVAILABLE:
     from torchvision import transforms
     from torchvision.datasets.mnist import MNIST
 else:

pl_examples/basic_examples/dali_image_classifier.py

@@ -23,9 +23,15 @@
 from torch.utils.data import random_split
 
 import pytorch_lightning as pl
-from pl_examples import _DALI_AVAILABLE, _DATASETS_PATH, _TORCHVISION_AVAILABLE, cli_lightning_logo
-
-if _TORCHVISION_AVAILABLE:
+from pl_examples import (
+    _DALI_AVAILABLE,
+    _DATASETS_PATH,
+    _TORCHVISION_AVAILABLE,
+    _TORCHVISION_MNIST_AVAILABLE,
+    cli_lightning_logo,
+)
+
+if _TORCHVISION_AVAILABLE and _TORCHVISION_MNIST_AVAILABLE:
     from torchvision import transforms
     from torchvision.datasets.mnist import MNIST
 else:

pl_examples/basic_examples/mnist_datamodule.py

@@ -17,10 +17,10 @@
 
 from torch.utils.data import DataLoader, random_split
 
-from pl_examples import _DATASETS_PATH, _TORCHVISION_AVAILABLE
+from pl_examples import _DATASETS_PATH, _TORCHVISION_AVAILABLE, _TORCHVISION_MNIST_AVAILABLE
 from pytorch_lightning import LightningDataModule
 
-if _TORCHVISION_AVAILABLE:
+if _TORCHVISION_AVAILABLE and _TORCHVISION_MNIST_AVAILABLE:
     from torchvision import transforms as transform_lib
     from torchvision.datasets import MNIST
 else: