CI: enable testing with coming PT 2.2

.azure/gpu-tests-fabric.yml

@@ -59,6 +59,9 @@ jobs:
         "Fabric | latest":
           image: "pytorchlightning/pytorch_lightning:base-cuda-py3.10-torch2.1-cuda12.1.0"
           PACKAGE_NAME: "fabric"
+        "Fabric | future":
+          image: "pytorchlightning/pytorch_lightning:base-cuda-py3.11-torch2.2-cuda12.1.0"
+          PACKAGE_NAME: "fabric"
         "Lightning | latest":
           image: "pytorchlightning/pytorch_lightning:base-cuda-py3.10-torch2.1-cuda12.1.0"
           PACKAGE_NAME: "lightning"
@@ -73,6 +76,10 @@ jobs:
           scope=$(python -c 'n = "$(PACKAGE_NAME)" ; print(dict(fabric="lightning_fabric").get(n, n))')
           echo "##vso[task.setvariable variable=COVERAGE_SOURCE]$scope"
         displayName: "set env. vars"
+      - bash: |
+          echo "##vso[task.setvariable variable=TORCH_URL]https://download.pytorch.org/whl/test/cu${CUDA_VERSION_MM}/torch_test.html"
+        condition: endsWith(variables['Agent.JobName'], 'future')
+        displayName: "set env. vars 4 future"
 
       - bash: |
           echo $(DEVICES)
@@ -99,7 +106,7 @@ jobs:
 
       - bash: |
           extra=$(python -c "print({'lightning': 'fabric-'}.get('$(PACKAGE_NAME)', ''))")
-          pip install -e ".[${extra}dev]" pytest-timeout -U --find-links ${TORCH_URL}
+          pip install -e ".[${extra}dev]" pytest-timeout -U --find-links="${TORCH_URL}"
         displayName: "Install package & dependencies"
 
       - bash: |

.azure/gpu-tests-pytorch.yml

@@ -51,6 +51,10 @@ jobs:
         "PyTorch | latest":
           image: "pytorchlightning/pytorch_lightning:base-cuda-py3.10-torch2.1-cuda12.1.0"
           PACKAGE_NAME: "pytorch"
+        "PyTorch | future":
+          # todo: failed to install `pygame` with py3.11
+          image: "pytorchlightning/pytorch_lightning:base-cuda-py3.10-torch2.2-cuda12.1.0"
+          PACKAGE_NAME: "pytorch"
         "Lightning | latest":
           image: "pytorchlightning/pytorch_lightning:base-cuda-py3.10-torch2.1-cuda12.1.0"
           PACKAGE_NAME: "lightning"
@@ -76,6 +80,10 @@ jobs:
           scope=$(python -c 'n = "$(PACKAGE_NAME)" ; print(dict(pytorch="pytorch_lightning").get(n, n))')
           echo "##vso[task.setvariable variable=COVERAGE_SOURCE]$scope"
         displayName: "set env. vars"
+      - bash: |
+          echo "##vso[task.setvariable variable=TORCH_URL]https://download.pytorch.org/whl/test/cu${CUDA_VERSION_MM}/torch_test.html"
+        condition: endsWith(variables['Agent.JobName'], 'future')
+        displayName: "set env. vars 4 future"
 
       - bash: |
           echo $(DEVICES)
@@ -109,7 +117,7 @@ jobs:
 
       - bash: |
           extra=$(python -c "print({'lightning': 'pytorch-'}.get('$(PACKAGE_NAME)', ''))")
-          pip install -e ".[${extra}dev]" -r requirements/_integrations/strategies.txt pytest-timeout -U --find-links ${TORCH_URL}
+          pip install -e ".[${extra}dev]" -r requirements/_integrations/strategies.txt pytest-timeout -U --find-links="${TORCH_URL}"
         displayName: "Install package & dependencies"
 
       - bash: pip uninstall -y lightning

.github/workflows/ci-tests-fabric.yml

@@ -45,13 +45,17 @@ jobs:
           - { os: "ubuntu-20.04", pkg-name: "lightning", python-version: "3.10", pytorch-version: "1.13" }
           - { os: "windows-2022", pkg-name: "lightning", python-version: "3.9", pytorch-version: "1.12" }
           - { os: "windows-2022", pkg-name: "lightning", python-version: "3.10", pytorch-version: "1.13" }
-          # only run PyTorch latest with Python recent
+          # only run PyTorch latest
           - { os: "macOS-11", pkg-name: "lightning", python-version: "3.10", pytorch-version: "2.0" }
           - { os: "ubuntu-20.04", pkg-name: "lightning", python-version: "3.10", pytorch-version: "2.0" }
           - { os: "windows-2022", pkg-name: "lightning", python-version: "3.10", pytorch-version: "2.0" }
           - { os: "macOS-11", pkg-name: "lightning", python-version: "3.11", pytorch-version: "2.1" }
           - { os: "ubuntu-20.04", pkg-name: "lightning", python-version: "3.11", pytorch-version: "2.1" }
           - { os: "windows-2022", pkg-name: "lightning", python-version: "3.11", pytorch-version: "2.1" }
+          # only run PyTorch future
+          - { os: "macOS-12", pkg-name: "lightning", python-version: "3.11", pytorch-version: "2.2" }
+          - { os: "ubuntu-22.04", pkg-name: "lightning", python-version: "3.11", pytorch-version: "2.2" }
+          - { os: "windows-2022", pkg-name: "lightning", python-version: "3.11", pytorch-version: "2.2" }
           # only run PyTorch latest with Python latest, use Fabric scope to limit dependency issues
           - { os: "macOS-12", pkg-name: "fabric", python-version: "3.11", pytorch-version: "2.0" }
           - { os: "ubuntu-22.04", pkg-name: "fabric", python-version: "3.11", pytorch-version: "2.0" }
@@ -128,7 +132,7 @@ jobs:
       - name: Env. variables
         run: |
           # Switch PyTorch URL
-          #python -c "print('TORCH_URL=' + str('${{env.TORCH_URL_TEST}}' if '${{ matrix.release }}' == 'pre' else '${{env.TORCH_URL_STABLE}}'))" >> $GITHUB_ENV
+          python -c "print('TORCH_URL=' + str('${{env.TORCH_URL_TEST}}' if '${{ matrix.pytorch-version }}' == '2.2' else '${{env.TORCH_URL_STABLE}}'))" >> $GITHUB_ENV
           # Switch coverage scope
           python -c "print('COVERAGE_SCOPE=' + str('lightning' if '${{matrix.pkg-name}}' == 'lightning' else 'lightning_fabric'))" >> $GITHUB_ENV
           # if you install mono-package set dependency only for this subpackage
@@ -157,7 +161,7 @@ jobs:
 
       - name: Testing Warnings
         working-directory: tests/tests_fabric
-        # needs to run outside of `pytest`
+        # needs to run outside `pytest`
         run: python utilities/test_warnings.py
 
       - name: Testing Fabric

.github/workflows/ci-tests-pytorch.yml

@@ -49,13 +49,17 @@ jobs:
           - { os: "ubuntu-20.04", pkg-name: "lightning", python-version: "3.10", pytorch-version: "1.13" }
           - { os: "windows-2022", pkg-name: "lightning", python-version: "3.9", pytorch-version: "1.12" }
           - { os: "windows-2022", pkg-name: "lightning", python-version: "3.10", pytorch-version: "1.13" }
-          # only run PyTorch latest with Python recent
+          # only run PyTorch latest
           - { os: "macOS-11", pkg-name: "lightning", python-version: "3.10", pytorch-version: "2.0" }
           - { os: "ubuntu-20.04", pkg-name: "lightning", python-version: "3.10", pytorch-version: "2.0" }
           - { os: "windows-2022", pkg-name: "lightning", python-version: "3.10", pytorch-version: "2.0" }
           - { os: "macOS-11", pkg-name: "lightning", python-version: "3.10", pytorch-version: "2.1" }
           - { os: "ubuntu-20.04", pkg-name: "lightning", python-version: "3.10", pytorch-version: "2.1" }
           - { os: "windows-2022", pkg-name: "lightning", python-version: "3.10", pytorch-version: "2.1" }
+          # only run PyTorch future
+          - { os: "macOS-12", pkg-name: "lightning", python-version: "3.10", pytorch-version: "2.2" }
+          - { os: "ubuntu-22.04", pkg-name: "lightning", python-version: "3.10", pytorch-version: "2.2" }
+          - { os: "windows-2022", pkg-name: "lightning", python-version: "3.10", pytorch-version: "2.2" }
           # only run PyTorch latest with Python latest, use PyTorch scope to limit dependency issues
           - { os: "macOS-12", pkg-name: "pytorch", python-version: "3.11", pytorch-version: "2.0" }
           - { os: "ubuntu-22.04", pkg-name: "pytorch", python-version: "3.11", pytorch-version: "2.0" }
@@ -134,7 +138,7 @@ jobs:
       - name: Env. variables
         run: |
           # Switch PyTorch URL
-          #python -c "print('TORCH_URL=' + str('${{env.TORCH_URL_TEST}}' if '${{ matrix.release }}' == 'pre' else '${{env.TORCH_URL_STABLE}}'))" >> $GITHUB_ENV
+          python -c "print('TORCH_URL=' + str('${{env.TORCH_URL_TEST}}' if '${{ matrix.pytorch-version }}' == '2.2' else '${{env.TORCH_URL_STABLE}}'))" >> $GITHUB_ENV
           # Switch coverage scope
           python -c "print('COVERAGE_SCOPE=' + str('lightning' if '${{matrix.pkg-name}}' == 'lightning' else 'pytorch_lightning'))" >> $GITHUB_ENV
           # if you install mono-package set dependency only for this subpackage
@@ -194,7 +198,7 @@ jobs:
 
       - name: Testing Warnings
         working-directory: tests/tests_pytorch
-        # needs to run outside of `pytest`
+        # needs to run outside `pytest`
         run: python utilities/test_warnings.py
 
       - name: Testing PyTorch

.github/workflows/docker-build.yml

@@ -75,7 +75,7 @@ jobs:
           tags = [f"latest-py{py_ver}-torch{pt_ver}-cuda{cuda_ver}"]
           if ver:
             tags += [f"{ver}-py{py_ver}-torch{pt_ver}-cuda{cuda_ver}"]
-          if py_ver == '3.10' and pt_ver == '2.0' and cuda_ver == '12.0.1':
+          if py_ver == '3.10' and pt_ver == '2.1' and cuda_ver == '12.1.0':
             tags += ["latest"]
 
           tags = [f"{repo}:{tag}" for tag in tags]
@@ -103,12 +103,16 @@ jobs:
       matrix:
         include:
           # These are the base images for PL release docker images,
-          # so include at least all of the combinations in release-dockers.yml.
+          # so include at least all the combinations in release-dockers.yml.
           - { python_version: "3.9", pytorch_version: "1.12", cuda_version: "11.7.1" }
           - { python_version: "3.9", pytorch_version: "1.13", cuda_version: "11.8.0" }
           - { python_version: "3.9", pytorch_version: "1.13", cuda_version: "12.0.1" }
           - { python_version: "3.10", pytorch_version: "2.0", cuda_version: "11.8.0" }
           - { python_version: "3.10", pytorch_version: "2.1", cuda_version: "12.1.0" }
+          - { python_version: "3.10", pytorch_version: "2.2", cuda_version: "12.1.0" }
+          - { python_version: "3.11", pytorch_version: "2.1", cuda_version: "12.1.0" }
+          - { python_version: "3.11", pytorch_version: "2.2", cuda_version: "12.1.0" }
+          # - { python_version: "3.12", pytorch_version: "2.2", cuda_version: "12.1.0" }  # todo: pending on `onnxruntime`
     steps:
       - uses: actions/checkout@v4
       - uses: docker/setup-buildx-action@v3

dockers/base-cuda/Dockerfile

@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-ARG UBUNTU_VERSION=20.04
+ARG UBUNTU_VERSION=22.04
 ARG CUDA_VERSION=11.7.1
 
 
@@ -38,7 +38,7 @@ RUN \
     # https://github.com/NVIDIA/nvidia-docker/issues/1631
     # https://github.com/NVIDIA/nvidia-docker/issues/1631#issuecomment-1264715214
     apt-get update && apt-get install -y wget && \
-    wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/3bf863cc.pub && \
+    wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/3bf863cc.pub && \
     mkdir -p /etc/apt/keyrings/ && mv 3bf863cc.pub /etc/apt/keyrings/ && \
     echo "deb [signed-by=/etc/apt/keyrings/3bf863cc.pub] https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/ /" /etc/apt/sources.list.d/cuda.list && \
     apt-get update -qq --fix-missing && \
@@ -82,9 +82,7 @@ COPY requirements/_integrations/ requirements/_integrations/
 ENV PYTHONPATH="/usr/lib/python${PYTHON_VERSION}/site-packages"
 
 RUN \
-    wget https://bootstrap.pypa.io/get-pip.py --progress=bar:force:noscroll --no-check-certificate && \
-    python${PYTHON_VERSION} get-pip.py && \
-    rm get-pip.py && \
+    curl https://bootstrap.pypa.io/get-pip.py | python${PYTHON_VERSION} && \
     # Disable cache \
     pip config set global.cache-dir false && \
     # set particular PyTorch version \
@@ -99,7 +97,8 @@ RUN \
       -r requirements/pytorch/extra.txt \
       -r requirements/pytorch/test.txt \
       -r requirements/pytorch/strategies.txt \
-      --find-links "https://download.pytorch.org/whl/cu${CUDA_VERSION_MM//'.'/''}/torch_stable.html"
+      --find-links="https://download.pytorch.org/whl/cu${CUDA_VERSION_MM//'.'/''}/torch_stable.html" \
+      --find-links="https://download.pytorch.org/whl/test/cu${CUDA_VERSION_MM//'.'/''}/torch_test.html"
 
 RUN \
     # Show what we have

requirements/app/app.txt

@@ -1,6 +1,6 @@
 lightning-cloud == 0.5.61  # Must be pinned to ensure compatibility
 packaging
-typing-extensions >=4.4.0, <4.8.0
+typing-extensions >=4.4.0, <4.10.0
 deepdiff >=5.7.0, <6.6.0
 fsspec[http] >=2022.5.0, <2023.11.0
 croniter >=1.3.0, <1.5.0  # strict; TODO: for now until we find something more robust.

requirements/fabric/base.txt

@@ -5,5 +5,5 @@ numpy >=1.17.2, <1.27.0
 torch >=1.12.0, <2.2.0
 fsspec[http] >=2022.5.0, <2023.11.0
 packaging >=20.0, <=23.1
-typing-extensions >=4.4.0, <4.8.0
+typing-extensions >=4.4.0, <4.10.0
 lightning-utilities >=0.8.0, <0.10.0

requirements/pytorch/base.txt

@@ -8,5 +8,5 @@ PyYAML >=5.4, <6.1.0
 fsspec[http] >=2022.5.0, <2023.11.0
 torchmetrics >=0.7.0, <1.3.0  # needed for using fixed compare_version
 packaging >=20.0, <=23.1
-typing-extensions >=4.4.0, <4.8.0
+typing-extensions >=4.4.0, <4.10.0
 lightning-utilities >=0.8.0, <0.10.0

requirements/pytorch/extra.txt

@@ -5,7 +5,7 @@
 matplotlib>3.1, <3.9.0
 omegaconf >=2.0.5, <2.4.0
 hydra-core >=1.0.5, <1.4.0
-jsonargparse[signatures] >=4.26.1, <4.27.0
+jsonargparse[signatures] >=4.26.1, <4.28.0
 rich >=12.3.0, <13.6.0
 tensorboardX >=2.2, <2.7.0  # min version is set by torch.onnx missing attribute
 bitsandbytes ==0.41.0  # strict

tests/tests_fabric/plugins/precision/test_amp_integration.py

@@ -12,10 +12,13 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """Integration tests for Automatic Mixed Precision (AMP) training."""
+import sys
+
 import pytest
 import torch
 import torch.nn as nn
 from lightning.fabric import Fabric, seed_everything
+from lightning.fabric.utilities.imports import _TORCH_GREATER_EQUAL_2_2
 
 from tests_fabric.helpers.runif import RunIf
 
@@ -37,6 +40,11 @@ def forward(self, x):
         return output
 
 
+@pytest.mark.xfail(
+    # https://github.com/pytorch/pytorch/issues/116056
+    sys.platform == "win32" and _TORCH_GREATER_EQUAL_2_2,
+    reason="Windows + DDP issue in PyTorch 2.2",
+)
 @pytest.mark.parametrize(
     ("accelerator", "precision", "expected_dtype"),
     [

tests/tests_fabric/strategies/launchers/test_multiprocessing_integration.py

@@ -11,10 +11,13 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+import sys
+
 import pytest
 import torch
 import torch.nn as nn
 from lightning.fabric import Fabric
+from lightning.fabric.utilities.imports import _TORCH_GREATER_EQUAL_2_2
 
 from tests_fabric.helpers.runif import RunIf
 
@@ -28,6 +31,11 @@ def __init__(self):
         self.register_buffer("buffer", torch.ones(3))
 
 
+@pytest.mark.xfail(
+    # https://github.com/pytorch/pytorch/issues/116056
+    sys.platform == "win32" and _TORCH_GREATER_EQUAL_2_2,
+    reason="Windows + DDP issue in PyTorch 2.2",
+)
 @pytest.mark.parametrize("strategy", ["ddp_spawn", pytest.param("ddp_fork", marks=RunIf(skip_windows=True))])
 def test_memory_sharing_disabled(strategy):
     """Test that the multiprocessing launcher disables memory sharing on model parameters and buffers to avoid race

tests/tests_fabric/strategies/test_ddp_integration.py

@@ -12,21 +12,27 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import os
+import sys
 from copy import deepcopy
 from unittest import mock
 from unittest.mock import Mock
 
 import pytest
 import torch
 from lightning.fabric import Fabric
-from lightning.fabric.utilities.imports import _TORCH_GREATER_EQUAL_2_0
+from lightning.fabric.utilities.imports import _TORCH_GREATER_EQUAL_2_0, _TORCH_GREATER_EQUAL_2_2
 from torch.nn.parallel.distributed import DistributedDataParallel
 
 from tests_fabric.helpers.runif import RunIf
 from tests_fabric.strategies.test_single_device import _run_test_clip_gradients
 from tests_fabric.test_fabric import BoringModel
 
 
+@pytest.mark.xfail(
+    # https://github.com/pytorch/pytorch/issues/116056
+    sys.platform == "win32" and _TORCH_GREATER_EQUAL_2_2,
+    reason="Windows + DDP issue in PyTorch 2.2",
+)
 @pytest.mark.parametrize(
     "accelerator",
     [

tests/tests_fabric/strategies/test_fsdp_integration.py

@@ -22,7 +22,10 @@
 from lightning.fabric import Fabric
 from lightning.fabric.plugins import FSDPPrecision
 from lightning.fabric.strategies import FSDPStrategy
-from lightning.fabric.utilities.imports import _TORCH_GREATER_EQUAL_2_0, _TORCH_GREATER_EQUAL_2_1
+from lightning.fabric.utilities.imports import (
+    _TORCH_GREATER_EQUAL_2_0,
+    _TORCH_GREATER_EQUAL_2_1,
+)
 from lightning.fabric.utilities.load import _load_distributed_checkpoint
 from lightning.fabric.wrappers import _FabricOptimizer
 from torch.distributed.fsdp import FlatParameter, FullyShardedDataParallel, OptimStateKeyType
@@ -560,7 +563,8 @@ def test_clip_gradients(clip_type, precision):
     optimizer.zero_grad()
 
 
-@RunIf(min_cuda_gpus=2, standalone=True, min_torch="2.1.0")
+# TODO: Support checkpoint consolidation with PyTorch >= 2.2
+@RunIf(min_cuda_gpus=2, standalone=True, min_torch="2.1.0", max_torch="2.2.0")
 def test_save_sharded_and_consolidate_and_load(tmp_path):
     """Test the consolidation of a FSDP-sharded checkpoint into a single file."""
 

tests/tests_fabric/utilities/test_distributed.py

@@ -1,5 +1,6 @@
 import functools
 import os
+import sys
 from functools import partial
 from pathlib import Path
 from unittest import mock
@@ -17,6 +18,7 @@
     _sync_ddp,
     is_shared_filesystem,
 )
+from lightning.fabric.utilities.imports import _TORCH_GREATER_EQUAL_2_2
 
 from tests_fabric.helpers.runif import RunIf
 
@@ -118,6 +120,11 @@ def test_collective_operations(devices, process):
     spawn_launch(process, devices)
 
 
+@pytest.mark.xfail(
+    # https://github.com/pytorch/pytorch/issues/116056
+    sys.platform == "win32" and _TORCH_GREATER_EQUAL_2_2,
+    reason="Windows + DDP issue in PyTorch 2.2",
+)
 @pytest.mark.flaky(reruns=3)  # flaky with "process 0 terminated with signal SIGABRT" (GLOO)
 def test_is_shared_filesystem(tmp_path, monkeypatch):
     # In the non-distributed case, every location is interpreted as 'shared'

tests/tests_fabric/utilities/test_spike.py

@@ -4,6 +4,7 @@
 import pytest
 import torch
 from lightning.fabric import Fabric
+from lightning.fabric.utilities.imports import _TORCH_GREATER_EQUAL_2_2
 from lightning.fabric.utilities.spike import _TORCHMETRICS_GREATER_EQUAL_1_0_0, SpikeDetection, TrainingSpikeException
 
 
@@ -28,6 +29,11 @@ def spike_detection_test(fabric, global_rank_spike, spike_value, should_raise):
             )
 
 
+@pytest.mark.xfail(
+    # https://github.com/pytorch/pytorch/issues/116056
+    sys.platform == "win32" and _TORCH_GREATER_EQUAL_2_2,
+    reason="Windows + DDP issue in PyTorch 2.2",
+)
 @pytest.mark.flaky(max_runs=3)
 @pytest.mark.parametrize(
     ("global_rank_spike", "num_devices", "spike_value", "finite_only"),