Merge branch 'master' into rich_progress_bar_vnum

Author: SkafteNicki

.github/CONTRIBUTING.md

@@ -113,14 +113,28 @@ ______________________________________________________________________
 
 To set up a local development environment, we recommend using `uv`, which can be installed following their [instructions](https://docs.astral.sh/uv/getting-started/installation/).
 
-Once `uv` has been installed, begin by cloning the repository:
+Once `uv` has been installed, begin by cloning the forked repository:
 
 ```bash
-git clone https://github.com/Lightning-AI/lightning.git
-cd lightning
+git clone https://github.com/{YOUR_GITHUB_USERNAME}/pytorch-lightning.git
+cd pytorch-lightning
 ```
 
-Once in root level of the repository, create a new virtual environment and install the project dependencies.
+> If you're using [Lightning Studio](https://lightning.ai) or already have your `uv venv` activated, you can quickly set up the project by running:
+
+```bash
+make setup
+```
+
+This will:
+
+- Install all required dependencies.
+- Perform an editable install of the `pytorch-lightning` project.
+- Install and configure `pre-commit`.
+
+#### Manual Setup (Optional)
+
+If you prefer more fine-grained control over the dependencies, you can set up the environment manually:
 
 ```bash
 uv venv

Makefile

@@ -1,4 +1,4 @@
-.PHONY: test clean docs
+.PHONY: test clean docs setup
 
 # to imitate SLURM set only single node
 export SLURM_LOCALID=0
@@ -7,6 +7,23 @@ export SPHINX_MOCK_REQUIREMENTS=1
 # install only Lightning Trainer packages
 export PACKAGE_NAME=pytorch
 
+setup:
+	uv pip install -r requirements.txt \
+	    -r requirements/pytorch/base.txt \
+	    -r requirements/pytorch/test.txt \
+	    -r requirements/pytorch/extra.txt \
+	    -r requirements/pytorch/strategies.txt \
+	    -r requirements/fabric/base.txt \
+	    -r requirements/fabric/test.txt \
+	    -r requirements/fabric/strategies.txt \
+	    -r requirements/typing.txt \
+	    -e ".[all]" \
+	    pre-commit
+	pre-commit install
+	@echo "-----------------------------"
+	@echo "✅ Environment setup complete. Ready to Contribute ⚡️!"
+
+
 clean:
 	# clean all temp runs
 	rm -rf $(shell find . -name "mlruns")

src/lightning/fabric/CHANGELOG.md

@@ -16,6 +16,10 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 -
 
+### Changed
+
+- Raise ValueError when seed is `out-of-bounds` or `cannot be cast to int` ([#21029](https://github.com/Lightning-AI/pytorch-lightning/pull/21029))
+
 
 ---
 

src/lightning/fabric/utilities/seed.py

@@ -27,7 +27,8 @@ def seed_everything(seed: Optional[int] = None, workers: bool = False, verbose:
     Args:
         seed: the integer value seed for global random state in Lightning.
             If ``None``, it will read the seed from ``PL_GLOBAL_SEED`` env variable. If ``None`` and the
-            ``PL_GLOBAL_SEED`` env variable is not set, then the seed defaults to 0.
+            ``PL_GLOBAL_SEED`` env variable is not set, then the seed defaults to 0. If seed is
+            not in bounds or cannot be cast to int, a ValueError is raised.
         workers: if set to ``True``, will properly configure all dataloaders passed to the
             Trainer with a ``worker_init_fn``. If the user already provides such a function
             for their dataloaders, setting this argument will have no influence. See also:
@@ -44,14 +45,12 @@ def seed_everything(seed: Optional[int] = None, workers: bool = False, verbose:
             try:
                 seed = int(env_seed)
             except ValueError:
-                seed = 0
-                rank_zero_warn(f"Invalid seed found: {repr(env_seed)}, seed set to {seed}")
+                raise ValueError(f"Invalid seed specified via PL_GLOBAL_SEED: {repr(env_seed)}")
     elif not isinstance(seed, int):
         seed = int(seed)
 
     if not (min_seed_value <= seed <= max_seed_value):
-        rank_zero_warn(f"{seed} is not in bounds, numpy accepts from {min_seed_value} to {max_seed_value}")
-        seed = 0
+        raise ValueError(f"{seed} is not in bounds, numpy accepts from {min_seed_value} to {max_seed_value}")
 
     if verbose:
         log.info(rank_prefixed_message(f"Seed set to {seed}", _get_rank()))

src/lightning/pytorch/CHANGELOG.md

@@ -10,7 +10,8 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 ### Added
 
--
+- Added support for general mappings being returned from `training_step` when using manual optimization ([#21011](https://github.com/Lightning-AI/pytorch-lightning/pull/21011))
+
 
 
 ### Changed

src/lightning/pytorch/loops/optimization/manual.py

@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 from collections import OrderedDict
+from collections.abc import Mapping
 from contextlib import suppress
 from dataclasses import dataclass, field
 from typing import Any
@@ -45,7 +46,7 @@ class ManualResult(OutputResult):
     @classmethod
     def from_training_step_output(cls, training_step_output: STEP_OUTPUT) -> "ManualResult":
         extra = {}
-        if isinstance(training_step_output, dict):
+        if isinstance(training_step_output, Mapping):
             extra = training_step_output.copy()
         elif isinstance(training_step_output, Tensor):
             extra = {"loss": training_step_output}

tests/tests_fabric/utilities/test_seed.py

@@ -47,19 +47,29 @@ def test_correct_seed_with_environment_variable():
 
 @mock.patch.dict(os.environ, {"PL_GLOBAL_SEED": "invalid"}, clear=True)
 def test_invalid_seed():
-    """Ensure that we still fix the seed even if an invalid seed is given."""
-    with pytest.warns(UserWarning, match="Invalid seed found"):
-        seed = seed_everything()
-    assert seed == 0
+    """Ensure that a ValueError is raised if an invalid seed is given."""
+    with pytest.raises(ValueError, match="Invalid seed specified"):
+        seed_everything()
 
 
 @mock.patch.dict(os.environ, {}, clear=True)
 @pytest.mark.parametrize("seed", [10e9, -10e9])
 def test_out_of_bounds_seed(seed):
-    """Ensure that we still fix the seed even if an out-of-bounds seed is given."""
-    with pytest.warns(UserWarning, match="is not in bounds"):
-        actual = seed_everything(seed)
-    assert actual == 0
+    """Ensure that a ValueError is raised if an out-of-bounds seed is given."""
+    with pytest.raises(ValueError, match="is not in bounds"):
+        seed_everything(seed)
+
+
+def test_seed_everything_accepts_valid_seed_argument():
+    """Ensure that seed_everything returns the provided valid seed."""
+    seed_value = 45
+    assert seed_everything(seed_value) == seed_value
+
+
+@mock.patch.dict(os.environ, {"PL_GLOBAL_SEED": "17"}, clear=True)
+def test_seed_everything_accepts_valid_seed_from_env():
+    """Ensure that seed_everything uses the valid seed from the PL_GLOBAL_SEED environment variable."""
+    assert seed_everything() == 17
 
 
 def test_reset_seed_no_op():

tests/tests_pytorch/trainer/optimization/test_manual_optimization.py

@@ -304,8 +304,36 @@ def on_train_epoch_end(self, *_, **__):
     trainer.fit(model)
 
 
+class CustomMapping(collections.abc.Mapping):
+    """A custom implementation of Mapping for testing purposes."""
+
+    def __init__(self, *args, **kwargs):
+        self._store = dict(*args, **kwargs)
+
+    def __getitem__(self, key):
+        return self._store[key]
+
+    def __iter__(self):
+        return iter(self._store)
+
+    def __len__(self):
+        return len(self._store)
+
+    def __repr__(self):
+        return f"{self.__class__.__name__}({self._store})"
+
+    def __copy__(self):
+        cls = self.__class__
+        new_obj = cls(self._store.copy())
+        return new_obj
+
+    def copy(self):
+        return self.__copy__()
+
+
 @RunIf(min_cuda_gpus=1)
-def test_multiple_optimizers_step(tmp_path):
+@pytest.mark.parametrize("dicttype", [dict, CustomMapping])
+def test_multiple_optimizers_step(tmp_path, dicttype):
     """Tests that `step` works with several optimizers."""
 
     class TestModel(ManualOptModel):
@@ -335,7 +363,7 @@ def training_step(self, batch, batch_idx):
             opt_b.step()
             opt_b.zero_grad()
 
-            return {"loss1": loss_1.detach(), "loss2": loss_2.detach()}
+            return dicttype(loss1=loss_1.detach(), loss2=loss_2.detach())
 
         # sister test: tests/plugins/test_amp_plugins.py::test_amp_gradient_unscale
         def on_after_backward(self) -> None: