Merge branch 'master' into lr-finder

Author: Borda

.actions/assistant.py

@@ -459,6 +459,25 @@ def generate_docker_tags(
         tags = [f"{docker_project}:{tag}" for tag in tags]
         print(",".join(tags))
 
+    @staticmethod
+    def prune_pytest_as_errors(
+        pyproject_toml: str = "pyproject.toml", errors: tuple = ("FutureWarning", "DeprecationWarning")
+    ) -> None:
+        """Prune pytest warnings as errors from the pyproject.toml file."""
+        import tomlkit
+
+        with open(pyproject_toml, encoding="utf-8") as fopen:
+            content = fopen.read()
+        pyproject = tomlkit.parse(content)
+        filterwarnings = pyproject.get("tool", {}).get("pytest", {}).get("ini_options", {}).get("filterwarnings", [])
+        if not filterwarnings:
+            return
+        filterwarnings = [wrn for wrn in filterwarnings if not any(f"error::{err}" in wrn for err in errors)]
+        pyproject["tool"]["pytest"]["ini_options"]["filterwarnings"] = filterwarnings
+
+        with open(pyproject_toml, "w", encoding="utf-8") as fopen:
+            fopen.write(tomlkit.dumps(pyproject))
+
 
 if __name__ == "__main__":
     import jsonargparse

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

@@ -99,11 +99,14 @@ jobs:
       - name: Set min. dependencies
         if: ${{ matrix.config.requires == 'oldest' }}
         run: |
-          cd requirements/fabric
           uv pip install -U "lightning-utilities[cli]"
-          python -m lightning_utilities.cli requirements set-oldest --req_files "['base.txt', 'strategies.txt', 'test.txt']"
+          python -m lightning_utilities.cli requirements set-oldest \
+            --req_files "['requirements/fabric/base.txt', 'requirements/fabric/strategies.txt', 'requirements/fabric/test.txt']"
           uv pip install "cython<3.0" wheel
           uv pip install "pyyaml==5.4" --no-build-isolation
+          # This script removes any line containing "error::FutureWarning" from pyproject.toml
+          uv pip install -r requirements/ci.txt
+          python .actions/assistant.py prune_pytest_as_errors
 
       - name: Adjust PyTorch versions in requirements files
         if: ${{ matrix.config.requires != 'oldest' }}

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

@@ -98,11 +98,14 @@ jobs:
       - name: Set min. dependencies
         if: ${{ matrix.config.requires == 'oldest' }}
         run: |
-          cd requirements/pytorch
           uv pip install -U "lightning-utilities[cli]"
-          python -m lightning_utilities.cli requirements set-oldest --req_files "['base.txt', 'extra.txt', 'strategies.txt', 'examples.txt', 'test.txt']"
+          python -m lightning_utilities.cli requirements set-oldest \
+            --req_files "['requirements/pytorch/base.txt', 'requirements/pytorch/extra.txt', 'requirements/pytorch/strategies.txt', 'requirements/pytorch/examples.txt', 'requirements/pytorch/test.txt']"
           uv pip install "cython<3.0" wheel
           uv pip install "pyyaml==5.4" --no-build-isolation
+          # This script removes any line containing "error::FutureWarning" from pyproject.toml
+          uv pip install -r requirements/ci.txt
+          python .actions/assistant.py prune_pytest_as_errors
 
       - name: Adjust PyTorch versions in requirements files
         if: ${{ matrix.config.requires != 'oldest' }}

.github/workflows/code-checks.yml

@@ -29,23 +29,32 @@ jobs:
     runs-on: ubuntu-22.04
     steps:
       - uses: actions/checkout@v5
-      - uses: actions/setup-python@v6
-        with:
-          python-version: "3.11"
 
-      - name: Mypy cache
-        uses: actions/cache@v4
+      - name: Install uv and set Python version
+        uses: astral-sh/setup-uv@v6
         with:
-          path: .mypy_cache
-          key: mypy-${{ hashFiles('requirements/typing.txt') }}
+          python-version: "3.11"
+          # 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 dependencies
         env:
           FREEZE_REQUIREMENTS: 1
         timeout-minutes: 20
         run: |
-          pip install -e '.[pytorch-all,fabric-all]' -r requirements/typing.txt
-          pip list
+          uv pip install '.[pytorch-all,fabric-all]' -r requirements/typing.txt
+          uv pip list
+
+      - name: mypy cache
+        uses: actions/cache@v4
+        with:
+          path: .mypy_cache
+          key: mypy-${{ hashFiles('requirements/typing.txt') }}
 
       - name: Check typing
         run: mypy
+
+      - name: Minimize uv cache
+        run: uv cache prune --ci

docs/source-pytorch/common/early_stopping.rst

@@ -1,6 +1,7 @@
 .. testsetup:: *
 
-    from lightning.pytorch.callbacks.early_stopping import EarlyStopping
+    from lightning.pytorch.callbacks.early_stopping import EarlyStopping, EarlyStoppingReason
+    from lightning.pytorch import Trainer, LightningModule
 
 .. _early_stopping:
 
@@ -71,6 +72,37 @@ Additional parameters that stop training at extreme points:
 - ``check_on_train_epoch_end``: When turned on, it checks the metric at the end of a training epoch. Use this only when you are monitoring any metric logged within
   training-specific hooks on epoch-level.
 
+After training completes, you can programmatically check why early stopping occurred using the ``stopping_reason``
+attribute, which returns an ``EarlyStoppingReason`` enum value.
+
+.. code-block:: python
+
+    from lightning.pytorch.callbacks import EarlyStopping
+    from lightning.pytorch.callbacks.early_stopping import EarlyStoppingReason
+
+    early_stopping = EarlyStopping(monitor="val_loss", patience=3)
+    trainer = Trainer(callbacks=[early_stopping])
+    trainer.fit(model)
+
+    # Check why training stopped
+    if early_stopping.stopping_reason == EarlyStoppingReason.PATIENCE_EXHAUSTED:
+        print("Training stopped due to patience exhaustion")
+    elif early_stopping.stopping_reason == EarlyStoppingReason.STOPPING_THRESHOLD:
+        print("Training stopped due to reaching stopping threshold")
+    elif early_stopping.stopping_reason == EarlyStoppingReason.NOT_STOPPED:
+        print("Training completed normally without early stopping")
+
+    # Access human-readable message
+    if early_stopping.stopping_reason_message:
+        print(f"Details: {early_stopping.stopping_reason_message}")
+
+The available stopping reasons are:
+
+- ``NOT_STOPPED``: Training completed normally without early stopping
+- ``STOPPING_THRESHOLD``: Training stopped because the monitored metric reached the stopping threshold
+- ``DIVERGENCE_THRESHOLD``: Training stopped because the monitored metric exceeded the divergence threshold
+- ``PATIENCE_EXHAUSTED``: Training stopped because the metric didn't improve for the specified patience
+- ``NON_FINITE_METRIC``: Training stopped because the monitored metric became NaN or infinite
 
 In case you need early stopping in a different part of training, subclass :class:`~lightning.pytorch.callbacks.early_stopping.EarlyStopping`
 and change where it is called:

pyproject.toml

@@ -182,7 +182,6 @@ filterwarnings = [
     # "error::DeprecationWarning",
     "error::FutureWarning",
     "ignore::FutureWarning:onnxscript", # Temporary ignore until onnxscript is updated
-    "ignore:The pynvml package is deprecated:FutureWarning", # Ignore pynvml deprecation warning, since it is not installed by PL directly
 ]
 xfail_strict = true
 junit_duration_report = "call"

requirements/ci.txt

@@ -6,3 +6,4 @@ importlib-metadata <9.0.0
 wget
 pkginfo ==1.12.1.2
 packaging <25.1
+tomlkit

requirements/fabric/test.txt

@@ -1,7 +1,7 @@
 coverage ==7.10.6
 numpy >=1.21.0, <1.27.0
 pytest ==8.4.2
-pytest-cov ==6.2.1
+pytest-cov ==6.3.0
 pytest-timeout ==2.4.0
 pytest-rerunfailures ==16.0.1
 pytest-random-order ==1.2.0

requirements/pytorch/examples.txt

@@ -3,5 +3,5 @@
 
 requests <2.33.0
 torchvision >=0.16.0, <0.24.0
-ipython[all] >=8.0.0, <9.0.0
+ipython[all] >=8.0.0, <10.0.0
 torchmetrics >=0.10.0, <1.9.0

requirements/pytorch/test.txt

@@ -1,6 +1,6 @@
 coverage ==7.10.6
 pytest ==8.4.2
-pytest-cov ==6.2.1
+pytest-cov ==6.3.0
 pytest-timeout ==2.4.0
 pytest-rerunfailures ==16.0.1
 pytest-random-order ==1.2.0