Merge branch 'master' into cpu_memory_accumulation_bug

Author: Borda

.actions/assistant.py

@@ -341,47 +341,6 @@ def create_mirror_package(source_dir: str, package_mapping: dict[str, str]) -> N
 
 
 class AssistantCLI:
-    @staticmethod
-    def requirements_prune_pkgs(packages: Sequence[str], req_files: Sequence[str] = REQUIREMENT_FILES_ALL) -> None:
-        """Remove some packages from given requirement files."""
-        if isinstance(req_files, str):
-            req_files = [req_files]
-        for req in req_files:
-            AssistantCLI._prune_packages(req, packages)
-
-    @staticmethod
-    def _prune_packages(req_file: str, packages: Sequence[str]) -> None:
-        """Remove some packages from given requirement files."""
-        path = Path(req_file)
-        assert path.exists()
-        text = path.read_text()
-        lines = text.splitlines()
-        final = []
-        for line in lines:
-            ln_ = line.strip()
-            if not ln_ or ln_.startswith("#"):
-                final.append(line)
-                continue
-            req = list(_parse_requirements([ln_]))[0]
-            if req.name not in packages:
-                final.append(line)
-        print(final)
-        path.write_text("\n".join(final) + "\n")
-
-    @staticmethod
-    def _replace_min(fname: str) -> None:
-        with open(fname, encoding="utf-8") as fopen:
-            req = fopen.read().replace(">=", "==")
-        with open(fname, "w", encoding="utf-8") as fwrite:
-            fwrite.write(req)
-
-    @staticmethod
-    def replace_oldest_ver(requirement_fnames: Sequence[str] = REQUIREMENT_FILES_ALL) -> None:
-        """Replace the min package version by fixed one."""
-        for fname in requirement_fnames:
-            print(fname)
-            AssistantCLI._replace_min(fname)
-
     @staticmethod
     def copy_replace_imports(
         source_dir: str,

.azure/gpu-benchmarks.yml

@@ -46,7 +46,7 @@ jobs:
     variables:
       DEVICES: $( python -c 'print("$(Agent.Name)".split("_")[-1])' )
     container:
-      image: "pytorchlightning/pytorch_lightning:base-cuda-py3.12-torch2.5-cuda12.1.0"
+      image: "pytorchlightning/pytorch_lightning:base-cuda12.6.3-py3.12-torch2.8"
       options: "--gpus=all --shm-size=32g"
     strategy:
       matrix:
@@ -96,7 +96,7 @@ jobs:
       - bash: python -m pytest parity_$(PACKAGE_NAME) -v --durations=0
         env:
           PL_RUNNING_BENCHMARKS: "1"
-          PL_RUN_CUDA_TESTS: "1"
+          RUN_ONLY_CUDA_TESTS: "1"
         workingDirectory: tests/
         displayName: "Testing: benchmarks"
 
@@ -105,7 +105,7 @@ jobs:
         # without succeeded this could run even if the job has already failed
         condition: and(succeeded(), eq(variables['PACKAGE_NAME'], 'fabric'))
         env:
-          PL_RUN_CUDA_TESTS: "1"
+          RUN_ONLY_CUDA_TESTS: "1"
           PL_RUN_STANDALONE_TESTS: "1"
         displayName: "Testing: fabric standalone tasks"
         timeoutInMinutes: "10"

.azure/gpu-tests-fabric.yml

@@ -48,7 +48,7 @@ jobs:
       DEVICES: $( python -c 'print("$(Agent.Name)".split("_")[-1])' )
       FREEZE_REQUIREMENTS: "1"
       PIP_CACHE_DIR: "/var/tmp/pip"
-      PL_RUN_CUDA_TESTS: "1"
+      RUN_ONLY_CUDA_TESTS: "1"
     container:
       image: $(image)
       # default shm size is 64m. Increase it to avoid:
@@ -57,16 +57,16 @@ jobs:
     strategy:
       matrix:
         "Fabric | oldest":
-          image: "pytorchlightning/pytorch_lightning:base-cuda-py3.10-torch2.1-cuda12.1.1"
+          image: "pytorchlightning/pytorch_lightning:base-cuda12.1.1-py3.10-torch2.1"
           PACKAGE_NAME: "fabric"
         "Fabric | latest":
-          image: "pytorchlightning/pytorch_lightning:base-cuda-py3.12-torch2.7-cuda12.6.3"
+          image: "pytorchlightning/pytorch_lightning:base-cuda12.6.3-py3.12-torch2.8"
           PACKAGE_NAME: "fabric"
         #"Fabric | future":
-        #  image: "pytorchlightning/pytorch_lightning:base-cuda-py3.12-torch2.7-cuda12.6.3"
+        #  image: "pytorchlightning/pytorch_lightning:base-cuda12.6.3-py3.12-torch2.7"
         #  PACKAGE_NAME: "fabric"
         "Lightning | latest":
-          image: "pytorchlightning/pytorch_lightning:base-cuda-py3.12-torch2.7-cuda12.6.3"
+          image: "pytorchlightning/pytorch_lightning:base-cuda12.6.3-py3.12-torch2.8"
           PACKAGE_NAME: "lightning"
     workspace:
       clean: all
@@ -78,8 +78,6 @@ jobs:
           echo "##vso[task.setvariable variable=TORCH_URL]https://download.pytorch.org/whl/cu${cuda_ver}/torch_stable.html"
           scope=$(python -c 'n = "$(PACKAGE_NAME)" ; print(dict(fabric="lightning_fabric").get(n, n))')
           echo "##vso[task.setvariable variable=COVERAGE_SOURCE]$scope"
-          python_ver=$(python -c "import sys; print(f'{sys.version_info.major}{sys.version_info.minor}')")
-          echo "##vso[task.setvariable variable=PYTHON_VERSION_MM]$python_ver"
         displayName: "set env. vars"
       - bash: |
           echo "##vso[task.setvariable variable=TORCH_URL]https://download.pytorch.org/whl/test/cu${CUDA_VERSION_MM}"
@@ -101,8 +99,16 @@ jobs:
         displayName: "Image info & NVIDIA"
 
       - bash: |
-          python .actions/assistant.py replace_oldest_ver
+          set -ex
           pip install "cython<3.0" wheel  # for compatibility
+          pip install -U "lightning-utilities[cli]"
+          cd requirements/fabric
+          # replace range by pin minimal requirements
+          python -m lightning_utilities.cli requirements set-oldest --req_files "['base.txt', 'strategies.txt']"
+          # drop deepspeed since it is not supported by our minimal Torch requirements
+          python -m lightning_utilities.cli requirements prune-pkgs --packages deepspeed --req_files strategies.txt
+          # uninstall deepspeed since some older docker images have it pre-installed
+          pip uninstall -y deepspeed
         condition: contains(variables['Agent.JobName'], 'oldest')
         displayName: "setting oldest dependencies"
 

.azure/gpu-tests-pytorch.yml

@@ -50,23 +50,23 @@ jobs:
     strategy:
       matrix:
         "PyTorch | oldest":
-          image: "pytorchlightning/pytorch_lightning:base-cuda-py3.10-torch2.1-cuda12.1.1"
+          image: "pytorchlightning/pytorch_lightning:base-cuda12.1.1-py3.10-torch2.1"
           PACKAGE_NAME: "pytorch"
         "PyTorch | latest":
-          image: "pytorchlightning/pytorch_lightning:base-cuda-py3.12-torch2.7-cuda12.6.3"
+          image: "pytorchlightning/pytorch_lightning:base-cuda12.6.3-py3.12-torch2.8"
           PACKAGE_NAME: "pytorch"
         #"PyTorch | future":
-        #  image: "pytorchlightning/pytorch_lightning:base-cuda-py3.12-torch2.7-cuda12.6.3"
+        #  image: "pytorchlightning/pytorch_lightning:base-cuda12.6.3-py3.12-torch2.7"
         #  PACKAGE_NAME: "pytorch"
         "Lightning | latest":
-          image: "pytorchlightning/pytorch_lightning:base-cuda-py3.12-torch2.7-cuda12.6.3"
+          image: "pytorchlightning/pytorch_lightning:base-cuda12.6.3-py3.12-torch2.8"
           PACKAGE_NAME: "lightning"
     pool: lit-rtx-3090
     variables:
       DEVICES: $( python -c 'print("$(Agent.Name)".split("_")[-1])' )
       FREEZE_REQUIREMENTS: "1"
       PIP_CACHE_DIR: "/var/tmp/pip"
-      PL_RUN_CUDA_TESTS: "1"
+      RUN_ONLY_CUDA_TESTS: "1"
     container:
       image: $(image)
       # default shm size is 64m. Increase it to avoid:
@@ -82,8 +82,6 @@ jobs:
           echo "##vso[task.setvariable variable=TORCH_URL]https://download.pytorch.org/whl/cu${cuda_ver}/torch_stable.html"
           scope=$(python -c 'n = "$(PACKAGE_NAME)" ; print(dict(pytorch="pytorch_lightning").get(n, n))')
           echo "##vso[task.setvariable variable=COVERAGE_SOURCE]$scope"
-          python_ver=$(python -c "import sys; print(f'{sys.version_info.major}{sys.version_info.minor}')")
-          echo "##vso[task.setvariable variable=PYTHON_VERSION_MM]$python_ver"
         displayName: "set env. vars"
       - bash: |
           echo "##vso[task.setvariable variable=TORCH_URL]https://download.pytorch.org/whl/test/cu${CUDA_VERSION_MM}"
@@ -105,8 +103,16 @@ jobs:
         displayName: "Image info & NVIDIA"
 
       - bash: |
-          python .actions/assistant.py replace_oldest_ver
+          set -ex
           pip install "cython<3.0" wheel  # for compatibility
+          pip install -U "lightning-utilities[cli]"
+          cd requirements/pytorch
+          # replace range by pin minimal requirements
+          python -m lightning_utilities.cli requirements set-oldest --req_files "['base.txt', 'extra.txt', 'strategies.txt', 'examples.txt']"
+          # drop deepspeed since it is not supported by our minimal Torch requirements
+          python -m lightning_utilities.cli requirements prune-pkgs --packages deepspeed --req_files strategies.txt
+          # uninstall deepspeed since some older docker images have it pre-installed
+          pip uninstall -y deepspeed
         condition: contains(variables['Agent.JobName'], 'oldest')
         displayName: "setting oldest dependencies"
 

.github/checkgroup.yml

@@ -48,6 +48,7 @@ subprojects:
       - "!**/*.md"
     checks:
       - "pytorch-lightning (GPUs) (testing Lightning | latest)"
+      - "pytorch-lightning (GPUs) (testing PyTorch | oldest)"
       - "pytorch-lightning (GPUs) (testing PyTorch | latest)"
 
   - id: "pytorch_lightning: Benchmarks"
@@ -174,6 +175,7 @@ subprojects:
       - "!*.md"
       - "!**/*.md"
     checks:
+      - "lightning-fabric (GPUs) (testing Fabric | oldest)"
       - "lightning-fabric (GPUs) (testing Fabric | latest)"
       - "lightning-fabric (GPUs) (testing Lightning | latest)"
 

.github/workflows/_build-packages.yml

@@ -25,7 +25,7 @@ jobs:
       matrix:
         pkg-name: ${{ fromJSON(inputs.pkg-names) }}
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
       - uses: actions/setup-python@v5
         with:
           python-version: "3.x"
@@ -51,7 +51,7 @@ jobs:
     needs: build-packages
     runs-on: ubuntu-22.04
     steps:
-      - uses: actions/download-artifact@v4
+      - uses: actions/download-artifact@v5
         with: # download all build artifacts
           pattern: ${{ inputs.artifact-name }}-*
           merge-multiple: true

.github/workflows/_legacy-checkpoints.yml

@@ -55,30 +55,34 @@ jobs:
     outputs:
       pl-version: ${{ steps.decide-version.outputs.pl-version }}
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
 
-      - uses: actions/setup-python@v5
+      - name: Install uv and set Python version
+        uses: astral-sh/setup-uv@v6
         with:
-          # Python version here needs to be supported by all PL versions listed in back-compatible-versions.txt.
           python-version: "3.9"
+          # TODO: Avoid activating environment like this
+          # see: https://github.com/astral-sh/setup-uv/tree/v6/?tab=readme-ov-file#activate-environment
+          activate-environment: true
+          enable-cache: true
 
       - name: Install PL from source
         env:
           PACKAGE_NAME: pytorch
           FREEZE_REQUIREMENTS: 1
         timeout-minutes: 20
-        run: pip install . --extra-index-url="${TORCH_URL}"
+        run: uv pip install . --extra-index-url="${TORCH_URL}"
         if: inputs.pl_version == ''
 
       - name: Install PL version
         timeout-minutes: 20
-        run: pip install "pytorch-lightning==${{ inputs.pl_version }}" --extra-index-url="${TORCH_URL}"
+        run: uv pip install "pytorch-lightning==${{ inputs.pl_version }}" --extra-index-url="${TORCH_URL}"
         if: inputs.pl_version != ''
 
       - name: Adjust tests -> PL
         if: ${{ matrix.pkg-name != 'lightning' }}
         run: |
-          pip install -q -r .actions/requirements.txt
+          uv pip install -q -r .actions/requirements.txt
           python .actions/assistant.py copy_replace_imports --source_dir="./tests" \
             --source_import="lightning.fabric,lightning.pytorch" \
             --target_import="lightning_fabric,pytorch_lightning"
@@ -115,7 +119,7 @@ jobs:
         # export to env bool if secrets.AWS_REGION is not empty
         run: echo "WITH_SECRETS=$([ -n '${{ secrets.AWS_REGION }}' ] && echo 1 || echo 0)" >> $GITHUB_ENV
 
-      - run: pip install -r requirements/ci.txt
+      - run: uv pip install -r requirements/ci.txt
       - name: Upload checkpoints to S3
         if: ${{ env.WITH_SECRETS == '1' }}
         working-directory: ${{ env.LEGACY_FOLDER }}
@@ -135,7 +139,7 @@ jobs:
     env:
       PL_VERSION: ${{ needs.create-legacy-ckpts.outputs.pl-version }}
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
         with:
           ref: master
 

.github/workflows/call-clear-cache.yml

@@ -23,18 +23,18 @@ on:
 jobs:
   cron-clear:
     if: github.event_name == 'schedule' || github.event_name == 'pull_request'
-    uses: Lightning-AI/utilities/.github/workflows/[email protected].0
+    uses: Lightning-AI/utilities/.github/workflows/[email protected].2
     with:
-      scripts-ref: v0.14.3
+      scripts-ref: v0.15.2
       dry-run: ${{ github.event_name == 'pull_request' }}
       pattern: "latest|docs"
       age-days: 7
 
   direct-clear:
     if: github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request'
-    uses: Lightning-AI/utilities/.github/workflows/[email protected].0
+    uses: Lightning-AI/utilities/.github/workflows/[email protected].2
     with:
-      scripts-ref: v0.14.3
+      scripts-ref: v0.15.2
       dry-run: ${{ github.event_name == 'pull_request' }}
       pattern: ${{ inputs.pattern || 'pypi_wheels' }} # setting str in case of PR / debugging
       age-days: ${{ fromJSON(inputs.age-days) || 0 }} # setting 0 in case of PR / debugging

.github/workflows/ci-pkg-install.yml

@@ -46,11 +46,11 @@ jobs:
         pkg-name: ["fabric", "pytorch", "lightning", "notset"]
         python-version: ["3.9", "3.11"]
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
       - uses: actions/setup-python@v5
         with:
           python-version: ${{ matrix.python-version }}
-      - uses: actions/download-artifact@v4
+      - uses: actions/download-artifact@v5
         with:
           name: dist-packages-${{ github.sha }}
           path: dist

.github/workflows/ci-schema.yml

@@ -8,7 +8,7 @@ on:
 
 jobs:
   check:
-    uses: Lightning-AI/utilities/.github/workflows/[email protected].0
+    uses: Lightning-AI/utilities/.github/workflows/[email protected].2
     with:
       # skip azure due to the wrong schema file by MSFT
       # https://github.com/Lightning-AI/lightning-flash/pull/1455#issuecomment-1244793607