Upgrade CI to PyTorch 1.13

.github/checkgroup.yml

@@ -1,7 +1,4 @@
 custom_service_name: "Lightning CI required checker"
-# For security reasons, configuration is only loaded from the repository's default branch,
-# changes made in pull requests from different branches or forks are ignored. This means that changes to this file
-# will only be used after they are merged.
 subprojects:
 
   # SECTION: pytorch_lightning
@@ -18,26 +15,25 @@ subprojects:
       - "setup.cfg"  # includes pytest config
       - ".actions/**"
     checks:
-      # Note: updates here should be applied to the lightning_lite group
-      - "pl-cpu (macOS-11, pytorch, 3.8, 1.10)"
-      - "pl-cpu (macOS-11, pytorch, 3.9, 1.11)"
-      - "pl-cpu (macOS-11, pytorch, 3.10, 1.12)"
-      - "pl-cpu (macOS-11, pytorch, 3.10, 1.13, pre)"
+      - "pl-cpu (macOS-11, pytorch, 3.8, 1.11)"
+      - "pl-cpu (macOS-11, pytorch, 3.9, 1.12)"
+      - "pl-cpu (macOS-11, pytorch, 3.10, 1.13)"
+      - "pl-cpu (macOS-11, pytorch, 3.8, 1.10, oldest)"
       - "pl-cpu (ubuntu-20.04, pytorch, 3.8, 1.10)"
       - "pl-cpu (ubuntu-20.04, pytorch, 3.9, 1.11)"
-      - "pl-cpu (ubuntu-20.04, pytorch, 3.10, 1.11)"
       - "pl-cpu (ubuntu-20.04, pytorch, 3.10, 1.12)"
-      - "pl-cpu (ubuntu-20.04, pytorch, 3.9, 1.13, pre)"
+      - "pl-cpu (ubuntu-20.04, pytorch, 3.10, 1.13)"
+      - "pl-cpu (ubuntu-20.04, pytorch, 3.7, 1.10, oldest)"
       - "pl-cpu (windows-2022, pytorch, 3.9, 1.11)"
-      - "pl-cpu (windows-2022, pytorch, 3.10, 1.11)"
       - "pl-cpu (windows-2022, pytorch, 3.10, 1.12)"
-      - "pl-cpu (windows-2022, pytorch, 3.8, 1.13, pre)"
-      - "pl-cpu (macOS-11, lightning, 3.10, 1.12)"
-      - "pl-cpu (ubuntu-20.04, lightning, 3.10, 1.12)"
-      - "pl-cpu (windows-2022, lightning, 3.10, 1.12)"
+      - "pl-cpu (windows-2022, pytorch, 3.10, 1.13)"
+      - "pl-cpu (windows-2022, pytorch, 3.7, 1.10, oldest)"
       - "pl-cpu (slow, macOS-11, pytorch, 3.7, 1.11)"
       - "pl-cpu (slow, ubuntu-20.04, pytorch, 3.7, 1.11)"
       - "pl-cpu (slow, windows-2022, pytorch, 3.7, 1.11)"
+      - "pl-cpu (macOS-11, lightning, 3.8, 1.13)"
+      - "pl-cpu (ubuntu-20.04, lightning, 3.8, 1.13)"
+      - "pl-cpu (windows-2022, lightning, 3.8, 1.13)"
 
   - id: "pytorch_lightning: Azure GPU"
     paths:
@@ -145,7 +141,7 @@ subprojects:
       - "build-pl (3.9, 1.10, 11.3.1)"
       - "build-pl (3.9, 1.11, 11.3.1)"
       - "build-pl (3.9, 1.12, 11.6.1)"
-      # TODO: add 1.13
+      - "build-pl (3.9, 1.13, 11.6.1)"
       - "build-xla (3.7, 1.12)"
 
   # SECTION: lightning_lite
@@ -159,20 +155,22 @@ subprojects:
       - ".github/workflows/ci-lite-tests.yml"
       - ".actions/**"
     checks:
-      - "lite-cpu (macOS-11, lite, 3.9, 1.11)"
-      - "lite-cpu (macOS-11, lite, 3.8, 1.10)"
-      - "lite-cpu (macOS-11, lite, 3.10, 1.12)"
-      - "lite-cpu (macOS-11, lite, 3.10, 1.13, pre)"
-      - "lite-cpu (ubuntu-20.04, lite, 3.8, 1.11)"
+      - "lite-cpu (macOS-11, lite, 3.8, 1.11)"
+      - "lite-cpu (macOS-11, lite, 3.9, 1.12)"
+      - "lite-cpu (macOS-11, lite, 3.10, 1.13)"
+      - "lite-cpu (macOS-11, lite, 3.7, 1.10, oldest)"
+      - "lite-cpu (ubuntu-20.04, lite, 3.8, 1.10)"
+      - "lite-cpu (ubuntu-20.04, lite, 3.9, 1.11)"
       - "lite-cpu (ubuntu-20.04, lite, 3.10, 1.12)"
-      - "lite-cpu (ubuntu-20.04, lite, 3.9, 1.13, pre)"
-      - "lite-cpu (windows-2022, lite, 3.9, 1.10)"
-      - "lite-cpu (windows-2022, lite, 3.10, 1.11)"
+      - "lite-cpu (ubuntu-20.04, lite, 3.10, 1.13)"
+      - "lite-cpu (ubuntu-20.04, lite, 3.7, 1.10, oldest)"
+      - "lite-cpu (windows-2022, lite, 3.9, 1.11)"
       - "lite-cpu (windows-2022, lite, 3.10, 1.12)"
-      - "lite-cpu (windows-2022, lite, 3.8, 1.13, pre)"
-      - "lite-cpu (macOS-11, lightning, 3.8, 1.12)"
-      - "lite-cpu (ubuntu-20.04, lightning, 3.8, 1.12)"
-      - "lite-cpu (windows-2022, lightning, 3.8, 1.12)"
+      - "lite-cpu (windows-2022, lite, 3.10, 1.13)"
+      - "lite-cpu (windows-2022, lite, 3.7, 1.10, oldest)"
+      - "lite-cpu (macOS-11, lightning, 3.8, 1.13)"
+      - "lite-cpu (ubuntu-20.04, lightning, 3.8, 1.13)"
+      - "lite-cpu (windows-2022, lightning, 3.8, 1.13)"
 
   - id: "lightning_lite: Azure GPU"
     paths:

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

@@ -36,25 +36,25 @@ jobs:
         include:
           # assign python and pytorch version combinations to operating systems (arbitrarily)
           # note: there's no distribution of torch==1.10 for Python>=3.10
-          - {os: "macOS-11", pkg-name: "lite", python-version: "3.9", pytorch-version: "1.11"}
-          - {os: "macOS-11", pkg-name: "lite", python-version: "3.8", pytorch-version: "1.10"}
-          - {os: "windows-2022", pkg-name: "lite", python-version: "3.10", pytorch-version: "1.11"}
-          - {os: "windows-2022", pkg-name: "lite", python-version: "3.9", pytorch-version: "1.10"}
-          - {os: "ubuntu-20.04", pkg-name: "lite", python-version: "3.8", pytorch-version: "1.11"}
-          # only run PyTorch latest with Python latest
-          - {os: "macOS-11", pkg-name: "lite", python-version: "3.10", pytorch-version: "1.12"}
+          - {os: "macOS-11", pkg-name: "lite", python-version: "3.8", pytorch-version: "1.11"}
+          - {os: "macOS-11", pkg-name: "lite", python-version: "3.9", pytorch-version: "1.12"}
+          - {os: "ubuntu-20.04", pkg-name: "lite", python-version: "3.8", pytorch-version: "1.10"}
+          - {os: "ubuntu-20.04", pkg-name: "lite", python-version: "3.9", pytorch-version: "1.11"}
           - {os: "ubuntu-20.04", pkg-name: "lite", python-version: "3.10", pytorch-version: "1.12"}
+          - {os: "windows-2022", pkg-name: "lite", python-version: "3.9", pytorch-version: "1.11"}
           - {os: "windows-2022", pkg-name: "lite", python-version: "3.10", pytorch-version: "1.12"}
+          # only run PyTorch latest with Python latest
+          - {os: "macOS-11", pkg-name: "lite", python-version: "3.10", pytorch-version: "1.13"}
+          - {os: "ubuntu-20.04", pkg-name: "lite", python-version: "3.10", pytorch-version: "1.13"}
+          - {os: "windows-2022", pkg-name: "lite", python-version: "3.10", pytorch-version: "1.13"}
           # "oldest" versions tests, only on minimum Python
-          # TODO: add back with 1.13
-          # release-candidate tests, mixed Python versions
-          - {os: "macOS-11", pkg-name: "lite", python-version: "3.10", pytorch-version: "1.13", release: "pre"}
-          - {os: "ubuntu-20.04", pkg-name: "lite", python-version: "3.9", pytorch-version: "1.13", release: "pre"}
-          - {os: "windows-2022", pkg-name: "lite", python-version: "3.8", pytorch-version: "1.13", release: "pre"}
+          - {os: "macOS-11", pkg-name: "lite", python-version: "3.7", pytorch-version: "1.10", requires: "oldest"}
+          - {os: "ubuntu-20.04", pkg-name: "lite", python-version: "3.7", pytorch-version: "1.10", requires: "oldest"}
+          - {os: "windows-2022", pkg-name: "lite", python-version: "3.7", pytorch-version: "1.10", requires: "oldest"}
           # "lightning" installs the monolithic package
-          - {os: "macOS-11", pkg-name: "lightning", python-version: "3.8", pytorch-version: "1.12"}
-          - {os: "ubuntu-20.04", pkg-name: "lightning", python-version: "3.8", pytorch-version: "1.12"}
-          - {os: "windows-2022", pkg-name: "lightning", python-version: "3.8", pytorch-version: "1.12"}
+          - {os: "macOS-11", pkg-name: "lightning", python-version: "3.8", pytorch-version: "1.13"}
+          - {os: "ubuntu-20.04", pkg-name: "lightning", python-version: "3.8", pytorch-version: "1.13"}
+          - {os: "windows-2022", pkg-name: "lightning", python-version: "3.8", pytorch-version: "1.13"}
 
     timeout-minutes: 15
 

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

@@ -40,33 +40,31 @@ jobs:
         include:
           # assign python and pytorch version combinations to operating systems (arbitrarily)
           # note: there's no distribution of torch==1.10 for Python>=3.10
-          - {os: "macOS-11", pkg-name: "pytorch", python-version: "3.9", pytorch-version: "1.11"}
-          - {os: "macOS-11", pkg-name: "pytorch", python-version: "3.8", pytorch-version: "1.10"}
+          - {os: "macOS-11", pkg-name: "pytorch", python-version: "3.8", pytorch-version: "1.11"}
+          - {os: "macOS-11", pkg-name: "pytorch", python-version: "3.9", pytorch-version: "1.12"}
           - {os: "ubuntu-20.04", pkg-name: "pytorch", python-version: "3.8", pytorch-version: "1.10"}
           - {os: "ubuntu-20.04", pkg-name: "pytorch", python-version: "3.9", pytorch-version: "1.11"}
-          - {os: "ubuntu-20.04", pkg-name: "pytorch", python-version: "3.10", pytorch-version: "1.11"}
-          - {os: "windows-2022", pkg-name: "pytorch", python-version: "3.8", pytorch-version: "1.10"}
-          - {os: "windows-2022", pkg-name: "pytorch", python-version: "3.9", pytorch-version: "1.11"}
-          - {os: "windows-2022", pkg-name: "pytorch", python-version: "3.10", pytorch-version: "1.11"}
-          # only run PyTorch latest with Python latest
-          - {os: "macOS-11", pkg-name: "pytorch", python-version: "3.10", pytorch-version: "1.12"}
-          - {os: "macOS-11", pkg-name: "lightning", python-version: "3.10", pytorch-version: "1.12"}
           - {os: "ubuntu-20.04", pkg-name: "pytorch", python-version: "3.10", pytorch-version: "1.12"}
-          - {os: "ubuntu-20.04", pkg-name: "lightning", python-version: "3.10", pytorch-version: "1.12"}
+          - {os: "windows-2022", pkg-name: "pytorch", python-version: "3.9", pytorch-version: "1.11"}
           - {os: "windows-2022", pkg-name: "pytorch", python-version: "3.10", pytorch-version: "1.12"}
-          - {os: "windows-2022", pkg-name: "lightning", python-version: "3.10", pytorch-version: "1.12"}
+          # only run PyTorch latest with Python latest
+          - {os: "macOS-11", pkg-name: "pytorch", python-version: "3.10", pytorch-version: "1.13"}
+          - {os: "ubuntu-20.04", pkg-name: "pytorch", python-version: "3.10", pytorch-version: "1.13"}
+          - {os: "windows-2022", pkg-name: "pytorch", python-version: "3.10", pytorch-version: "1.13"}
           # "oldest" versions tests, only on minimum Python
-          # TODO: add back with 1.13
-          # release-candidate tests, mixed Python versions
-          - {os: "macOS-11", pkg-name: "pytorch", python-version: "3.10", pytorch-version: "1.13", release: "pre"}
-          - {os: "ubuntu-20.04", pkg-name: "pytorch", python-version: "3.9", pytorch-version: "1.13", release: "pre"}
-          - {os: "windows-2022", pkg-name: "pytorch", python-version: "3.8", pytorch-version: "1.13", release: "pre"}
+          - {os: "macOS-11", pkg-name: "pytorch", python-version: "3.8", pytorch-version: "1.10", requires: "oldest"}  # 3.7 hangs
+          - {os: "ubuntu-20.04", pkg-name: "pytorch", python-version: "3.7", pytorch-version: "1.10", requires: "oldest"}
+          - {os: "windows-2022", pkg-name: "pytorch", python-version: "3.7", pytorch-version: "1.10", requires: "oldest"}
           # run test under SLOW label
           - {type: "slow", os: "macOS-11", pkg-name: "pytorch", python-version: "3.7", pytorch-version: "1.11"}
           - {type: "slow", os: "ubuntu-20.04", pkg-name: "pytorch", python-version: "3.7", pytorch-version: "1.11"}
           - {type: "slow", os: "windows-2022", pkg-name: "pytorch", python-version: "3.7", pytorch-version: "1.11"}
+          # "lightning" installs the monolithic package
+          - {os: "macOS-11", pkg-name: "lightning", python-version: "3.8", pytorch-version: "1.13"}
+          - {os: "ubuntu-20.04", pkg-name: "lightning", python-version: "3.8", pytorch-version: "1.13"}
+          - {os: "windows-2022", pkg-name: "lightning", python-version: "3.8", pytorch-version: "1.13"}
 
-    timeout-minutes: 60
+    timeout-minutes: 70  # tests with macOS-11, py3.7 oldest takes much longer then expected
 
     steps:
     - uses: actions/checkout@v3

requirements/lite/base.txt

@@ -2,7 +2,7 @@
 #  in case you want to preserve/enforce restrictions on the latest compatible version, add "strict" as an in-line comment
 
 numpy>=1.17.2, <1.23.1
-torch>=1.10.*, <1.13.0
+torch>=1.10.*, <=1.13.0
 fsspec[http]>2021.06.0, <2022.6.0
 packaging>=17.0, <=21.3
 typing-extensions>=4.0.0, <=4.4.0

requirements/pytorch/adjust-versions.py

@@ -5,8 +5,8 @@
 
 # IMPORTANT: this list needs to be sorted in reverse
 VERSIONS = [
-    dict(torch="1.13.0", torchvision="0.14.0"),  # RC
-    dict(torch="1.12.1", torchvision="0.13.1"),  # stable
+    dict(torch="1.13.0", torchvision="0.14.0"),  # stable
+    dict(torch="1.12.1", torchvision="0.13.1"),
     dict(torch="1.12.0", torchvision="0.13.0"),
     dict(torch="1.11.0", torchvision="0.12.0"),
     dict(torch="1.10.2", torchvision="0.11.3"),

requirements/pytorch/base.txt

@@ -2,7 +2,7 @@
 #  in case you want to preserve/enforce restrictions on the latest compatible version, add "strict" as an in-line comment
 
 numpy>=1.17.2, <1.23.1
-torch>=1.10.*, <1.13.0
+torch>=1.10.*, <=1.13.0
 tqdm>=4.57.0, <4.65.0
 PyYAML>=5.4, <=6.0
 fsspec[http]>2021.06.0, <2022.8.0

requirements/pytorch/check-avail-extras.py

@@ -1,5 +1,6 @@
-import hydra  # noqa: F401
-import jsonargparse  # noqa: F401
-import matplotlib  # noqa: F401
-import omegaconf  # noqa: F401
-import rich  # noqa: F401
+if __name__ == "__main__":
+    import hydra  # noqa: F401
+    import jsonargparse  # noqa: F401
+    import matplotlib  # noqa: F401
+    import omegaconf  # noqa: F401
+    import rich  # noqa: F401

requirements/pytorch/check-avail-strategies.py

@@ -1,7 +1,8 @@
-import bagua  # noqa: F401
-import deepspeed  # noqa: F401
-import fairscale  # noqa: F401
-import horovod.torch
+if __name__ == "__main__":
+    import bagua  # noqa: F401
+    import deepspeed  # noqa: F401
+    import fairscale  # noqa: F401
+    import horovod.torch
 
-# returns an error code
-assert horovod.torch.nccl_built()
+    # returns an error code
+    assert horovod.torch.nccl_built()

src/lightning_app/cli/app-template/tests/requirements.txt

@@ -1,8 +1,8 @@
 coverage
 codecov>=2.1
-pytest>=3.0.5
+pytest>=5.0.0
 pytest-cov
 pytest-flake8
 flake8
 check-manifest
-twine==1.13.0
+twine==4.0.1

src/lightning_app/cli/component-template/tests/requirements.txt

@@ -1,8 +1,8 @@
 coverage
 codecov>=2.1
-pytest>=3.0.5
+pytest>=5.0.0
 pytest-cov
 pytest-flake8
 flake8
 check-manifest
-twine==1.13.0
+twine==4.0.1

tests/tests_lite/plugins/precision/test_native_amp.py

@@ -41,19 +41,21 @@ def test_native_amp_precision_bf16_min_torch():
 
 
 def test_native_amp_precision_forward_context():
-    """Test to ensure that the context manager correctly is set to CPU + bfloat16."""
+    """Test to ensure that the context manager correctly is set to bfloat16 on CPU and CUDA."""
     precision = NativeMixedPrecision(precision=16, device="cuda")
     assert precision.device == "cuda"
     assert isinstance(precision.scaler, torch.cuda.amp.GradScaler)
     assert torch.get_default_dtype() == torch.float32
     with precision.forward_context():
-        assert torch.get_autocast_gpu_dtype() == torch.float16
+        # check with str due to a bug upstream: https://github.com/pytorch/pytorch/issues/65786
+        assert str(torch.get_autocast_gpu_dtype()) in ("torch.float16", "torch.half")
 
     precision = NativeMixedPrecision(precision="bf16", device="cpu")
     assert precision.device == "cpu"
     assert precision.scaler is None
     with precision.forward_context():
-        assert torch.get_autocast_cpu_dtype() == torch.bfloat16
+        # check with str due to a bug upstream: https://github.com/pytorch/pytorch/issues/65786
+        assert str(torch.get_autocast_cpu_dtype()) == str(torch.bfloat16)
 
     context_manager = precision._autocast_context_manager()
     assert isinstance(context_manager, torch.autocast)