Merge branch 'master' into ci/bump-pt-2.6

Borda · web-flow · commit d54ab0bf207e · 2025-02-15T00:01:46.000+01:00
diff --git a/.github/CONTRIBUTING.md b/.github/CONTRIBUTING.md
@@ -189,7 +189,7 @@ We welcome any useful contribution! For your convenience here's a recommended wo
 #### How can I help/contribute?
 
 All types of contributions are welcome - reporting bugs, fixing documentation, adding test cases, solving issues, and preparing bug fixes.
-To get started with code contributions, look for issues marked with the label [good first issue](https://github.com/Lightning-AI/lightning/issues?q=is%3Aopen+is%3Aissue+label%3A%22good+first+issue%22) or chose something close to your domain with the label [help wanted](https://github.com/Lightning-AI/lightning/issues?q=is%3Aopen+is%3Aissue+label%3A%22help+wanted%22). Before coding, make sure that the issue description is clear and comment on the issue so that we can assign it to you (or simply self-assign if you can).
+To get started with code contributions, look for issues marked with the label [good first issue](https://github.com/Lightning-AI/pytorch-lightning/issues?q=is%3Aopen+is%3Aissue+label%3A%22good+first+issue%22) or chose something close to your domain with the label [help wanted](https://github.com/Lightning-AI/pytorch-lightning/issues?q=is%3Aopen+is%3Aissue+label%3A%22help+wanted%22). Before coding, make sure that the issue description is clear and comment on the issue so that we can assign it to you (or simply self-assign if you can).
 
 #### Is there a recommendation for branch names?
 
diff --git a/docs/source-fabric/_templates/theme_variables.jinja b/docs/source-fabric/_templates/theme_variables.jinja
@@ -1,6 +1,6 @@
 {%- set external_urls = {
   'github': 'https://github.com/Lightning-AI/lightning',
-  'github_issues': 'https://github.com/Lightning-AI/lightning/issues',
+  'github_issues': 'https://github.com/Lightning-AI/pytorch-lightning/issues',
   'contributing': 'https://github.com/Lightning-AI/lightning/blob/master/.github/CONTRIBUTING.md',
   'governance': 'https://lightning.ai/docs/pytorch/latest/community/governance.html',
   'docs': 'https://lightning.ai/docs/fabric/',
diff --git a/docs/source-fabric/links.rst b/docs/source-fabric/links.rst
@@ -1,3 +1,3 @@
-.. _PyTorchJob: https://www.kubeflow.org/docs/components/training/pytorch/
+.. _PyTorchJob: https://www.kubeflow.org/docs/components/trainer/legacy-v1/user-guides/pytorch/
 .. _Kubeflow: https://www.kubeflow.org
 .. _Trainer: https://lightning.ai/docs/pytorch/stable/common/trainer.html
diff --git a/docs/source-pytorch/_templates/theme_variables.jinja b/docs/source-pytorch/_templates/theme_variables.jinja
@@ -1,6 +1,6 @@
 {%- set external_urls = {
   'github': 'https://github.com/Lightning-AI/lightning',
-  'github_issues': 'https://github.com/Lightning-AI/lightning/issues',
+  'github_issues': 'https://github.com/Lightning-AI/pytorch-lightning/issues',
   'contributing': 'https://github.com/Lightning-AI/lightning/blob/master/.github/CONTRIBUTING.md',
   'governance': 'https://lightning.ai/docs/pytorch/latest/community/governance.html',
   'docs': 'https://lightning.ai/docs/pytorch/latest/',
diff --git a/docs/source-pytorch/accelerators/accelerator_prepare.rst b/docs/source-pytorch/accelerators/accelerator_prepare.rst
@@ -123,7 +123,7 @@ It is possible to perform some computation manually and log the reduced result o
 
         # When you call `self.log` only on rank 0, don't forget to add
         # `rank_zero_only=True` to avoid deadlocks on synchronization.
-        # Caveat: monitoring this is unimplemented, see https://github.com/Lightning-AI/lightning/issues/15852
+        # Caveat: monitoring this is unimplemented, see https://github.com/Lightning-AI/pytorch-lightning/issues/15852
         if self.trainer.is_global_zero:
             self.log("my_reduced_metric", mean, rank_zero_only=True)
 
diff --git a/docs/source-pytorch/accelerators/gpu_intermediate.rst b/docs/source-pytorch/accelerators/gpu_intermediate.rst
@@ -25,10 +25,6 @@ Lightning supports multiple ways of doing distributed training.
 .. note::
     If you request multiple GPUs or nodes without setting a strategy, DDP will be automatically used.
 
-For a deeper understanding of what Lightning is doing, feel free to read this
-`guide <https://towardsdatascience.com/9-tips-for-training-lightning-fast-neural-networks-in-pytorch-8e63a502f565>`_.
-
-
 ----
 
 
diff --git a/docs/source-pytorch/advanced/ddp_optimizations.rst b/docs/source-pytorch/advanced/ddp_optimizations.rst
@@ -58,7 +58,7 @@ On a Multi-Node Cluster, Set NCCL Parameters
 ********************************************
 
 `NCCL <https://developer.nvidia.com/nccl>`__ is the NVIDIA Collective Communications Library that is used by PyTorch to handle communication across nodes and GPUs.
-There are reported benefits in terms of speedups when adjusting NCCL parameters as seen in this `issue <https://github.com/Lightning-AI/lightning/issues/7179>`__.
+There are reported benefits in terms of speedups when adjusting NCCL parameters as seen in this `issue <https://github.com/Lightning-AI/pytorch-lightning/issues/7179>`__.
 In the issue, we see a 30% speed improvement when training the Transformer XLM-RoBERTa and a 15% improvement in training with Detectron2.
 NCCL parameters can be adjusted via environment variables.
 
diff --git a/docs/source-pytorch/advanced/model_parallel/deepspeed.rst b/docs/source-pytorch/advanced/model_parallel/deepspeed.rst
@@ -319,7 +319,7 @@ Additionally, DeepSpeed supports offloading to NVMe drives for even larger model
     )
     trainer.fit(model)
 
-When offloading to NVMe you may notice that the speed is slow. There are parameters that need to be tuned based on the drives that you are using. Running the `aio_bench_perf_sweep.py <https://github.com/microsoft/DeepSpeed/blob/master/csrc/aio/py_test/aio_bench_perf_sweep.py>`__ script can help you to find optimum parameters. See the `issue <https://github.com/microsoft/DeepSpeed/issues/998>`__ for more information on how to parse the information.
+When offloading to NVMe you may notice that the speed is slow. There are parameters that need to be tuned based on the drives that you are using. Running the `aio_bench_perf_sweep.py <https://github.com/microsoft/DeepSpeed/blob/master/csrc/aio/py_test/aio_bench_perf_sweep.py>`__ script can help you to find optimum parameters. See the `issue <https://github.com/deepspeedai/DeepSpeed/issues/998>`__ for more information on how to parse the information.
 
 .. _deepspeed-activation-checkpointing:
 
diff --git a/docs/source-pytorch/data/alternatives.rst b/docs/source-pytorch/data/alternatives.rst
@@ -90,7 +90,7 @@ the desired GPU in your pipeline. When moving data to a specific device, you can
 WebDataset
 ^^^^^^^^^^
 
-The `WebDataset <https://webdataset.github.io/webdataset>`__ makes it easy to write I/O pipelines for large datasets.
+The `WebDataset <https://github.com/webdataset/webdataset>`__ makes it easy to write I/O pipelines for large datasets.
 Datasets can be stored locally or in the cloud. ``WebDataset`` is just an instance of a standard IterableDataset.
 The webdataset library contains a small wrapper (``WebLoader``) that adds a fluid interface to the DataLoader (and is otherwise identical).
 
diff --git a/docs/source-pytorch/data/iterables.rst b/docs/source-pytorch/data/iterables.rst
@@ -50,7 +50,7 @@ To choose a different mode, you can use the :class:`~lightning.pytorch.utilities
 
 
 Currently, the ``trainer.predict`` method only supports the ``"sequential"`` mode, while ``trainer.fit`` method does not support it.
-Support for this feature is tracked in this `issue <https://github.com/Lightning-AI/lightning/issues/16830>`__.
+Support for this feature is tracked in this `issue <https://github.com/Lightning-AI/pytorch-lightning/issues/16830>`__.
 
 Note that when using the ``"sequential"`` mode, you need to add an additional argument ``dataloader_idx`` to some specific hooks.
 Lightning will `raise an error <https://github.com/Lightning-AI/lightning/pull/16837>`__ informing you of this requirement.
diff --git a/docs/source-pytorch/links.rst b/docs/source-pytorch/links.rst
@@ -1,2 +1,2 @@
-.. _PyTorchJob: https://www.kubeflow.org/docs/components/training/pytorch/
+.. _PyTorchJob: https://www.kubeflow.org/docs/components/trainer/legacy-v1/user-guides/pytorch/
 .. _Kubeflow: https://www.kubeflow.org
diff --git a/docs/source-pytorch/versioning.rst b/docs/source-pytorch/versioning.rst
@@ -61,8 +61,8 @@ For API removal, renaming or other forms of backwards-incompatible changes, the
 #. From that version onward, the deprecation warning gets converted into a helpful error, which will remain until next major release.
 
 This policy is not strict. Shorter or longer deprecation cycles may apply to some cases.
-For example, in the past DDP2 was removed without a deprecation process because the feature was broken and unusable beyond fixing as discussed in `#12584 <https://github.com/Lightning-AI/lightning/issues/12584>`_.
-Also, `#10410 <https://github.com/Lightning-AI/lightning/issues/10410>`_ is an example that a longer deprecation applied to. We deprecated the accelerator arguments, such as ``Trainer(gpus=...)``, in 1.7, however, because the APIs were so core that they would impact almost all use cases, we decided not to introduce the breaking change until 2.0.
+For example, in the past DDP2 was removed without a deprecation process because the feature was broken and unusable beyond fixing as discussed in `#12584 <https://github.com/Lightning-AI/pytorch-lightning/issues/12584>`_.
+Also, `#10410 <https://github.com/Lightning-AI/pytorch-lightning/issues/10410>`_ is an example that a longer deprecation applied to. We deprecated the accelerator arguments, such as ``Trainer(gpus=...)``, in 1.7, however, because the APIs were so core that they would impact almost all use cases, we decided not to introduce the breaking change until 2.0.
 
 Compatibility matrix
 ********************
diff --git a/src/lightning/__setup__.py b/src/lightning/__setup__.py
@@ -104,7 +104,7 @@ def _setup_args() -> dict[str, Any]:
         "install_requires": install_requires,
         "extras_require": _prepare_extras(),
         "project_urls": {
-            "Bug Tracker": "https://github.com/Lightning-AI/lightning/issues",
+            "Bug Tracker": "https://github.com/Lightning-AI/pytorch-lightning/issues",
             "Documentation": "https://lightning.ai/lightning-docs",
             "Source Code": "https://github.com/Lightning-AI/lightning",
         },
diff --git a/src/lightning/fabric/CHANGELOG.md b/src/lightning/fabric/CHANGELOG.md
@@ -337,7 +337,7 @@ Removed legacy supoport for `lightning run model`. Use `fabric run` instead. ([#
 ### Fixed
 
 - Fixed computing the next version folder in `CSVLogger` ([#17139](https://github.com/Lightning-AI/lightning/pull/17139))
-- Fixed inconsistent settings for FSDP Precision ([#17670](https://github.com/Lightning-AI/lightning/issues/17670))
+- Fixed inconsistent settings for FSDP Precision ([#17670](https://github.com/Lightning-AI/pytorch-lightning/issues/17670))
 
 
 ## [2.0.2] - 2023-04-24
diff --git a/src/lightning/fabric/plugins/environments/kubeflow.py b/src/lightning/fabric/plugins/environments/kubeflow.py
@@ -28,7 +28,7 @@ class KubeflowEnvironment(ClusterEnvironment):
     This environment, unlike others, does not get auto-detected and needs to be passed to the Fabric/Trainer
     constructor manually.
 
-    .. _PyTorchJob: https://www.kubeflow.org/docs/components/training/pytorch/
+    .. _PyTorchJob: https://www.kubeflow.org/docs/components/trainer/legacy-v1/user-guides/pytorch/
     .. _Kubeflow: https://www.kubeflow.org
 
     """
diff --git a/src/lightning/fabric/strategies/launchers/multiprocessing.py b/src/lightning/fabric/strategies/launchers/multiprocessing.py
@@ -78,7 +78,7 @@ def __init__(
     def is_interactive_compatible(self) -> bool:
         # The start method 'spawn' is not supported in interactive environments
         # The start method 'fork' is the only one supported in Jupyter environments, with constraints around CUDA
-        # initialization. For more context, see https://github.com/Lightning-AI/lightning/issues/7550
+        # initialization. For more context, see https://github.com/Lightning-AI/pytorch-lightning/issues/7550
         return self._start_method == "fork"
 
     @override
diff --git a/src/lightning/fabric/strategies/launchers/subprocess_script.py b/src/lightning/fabric/strategies/launchers/subprocess_script.py
@@ -156,7 +156,7 @@ def _check_can_spawn_children(self) -> None:
 
 
 def _basic_subprocess_cmd() -> Sequence[str]:
-    import __main__  # local import to avoid https://github.com/Lightning-AI/lightning/issues/15218
+    import __main__  # local import to avoid https://github.com/Lightning-AI/pytorch-lightning/issues/15218
 
     if __main__.__spec__ is None:  # pragma: no-cover
         return [sys.executable, os.path.abspath(sys.argv[0])] + sys.argv[1:]
@@ -167,7 +167,7 @@ def _hydra_subprocess_cmd(local_rank: int) -> tuple[Sequence[str], str]:
     from hydra.core.hydra_config import HydraConfig
     from hydra.utils import get_original_cwd, to_absolute_path
 
-    import __main__  # local import to avoid https://github.com/Lightning-AI/lightning/issues/15218
+    import __main__  # local import to avoid https://github.com/Lightning-AI/pytorch-lightning/issues/15218
 
     # when user is using hydra find the absolute path
     if __main__.__spec__ is None:  # pragma: no-cover
diff --git a/src/lightning/pytorch/CHANGELOG.md b/src/lightning/pytorch/CHANGELOG.md
@@ -199,27 +199,27 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 - Fixed handling checkpoint dirpath suffix in NeptuneLogger ([#18863](https://github.com/Lightning-AI/lightning/pull/18863))
 - Fixed an edge case where `ModelCheckpoint` would alternate between versioned and unversioned filename ([#19064](https://github.com/Lightning-AI/lightning/pull/19064))
 - Fixed broadcast at initialization in `MPIEnvironment` ([#19074](https://github.com/Lightning-AI/lightning/pull/19074))
-- Fixed the tensor conversion in `self.log` to respect the default dtype ([#19046](https://github.com/Lightning-AI/lightning/issues/19046))
+- Fixed the tensor conversion in `self.log` to respect the default dtype ([#19046](https://github.com/Lightning-AI/pytorch-lightning/issues/19046))
 
 
 ## [2.1.2] - 2023-11-15
 
 ### Fixed
 
-- Fixed an issue causing permission errors on Windows when attempting to create a symlink for the "last" checkpoint ([#18942](https://github.com/Lightning-AI/lightning/issues/18942))
-- Fixed an issue where Metric instances from `torchmetrics` wouldn't get moved to the device when using FSDP ([#18954](https://github.com/Lightning-AI/lightning/issues/18954))
-- Fixed an issue preventing the user to `Trainer.save_checkpoint()` an FSDP model when `Trainer.test/validate/predict()` ran after `Trainer.fit()` ([#18992](https://github.com/Lightning-AI/lightning/issues/18992))
+- Fixed an issue causing permission errors on Windows when attempting to create a symlink for the "last" checkpoint ([#18942](https://github.com/Lightning-AI/pytorch-lightning/issues/18942))
+- Fixed an issue where Metric instances from `torchmetrics` wouldn't get moved to the device when using FSDP ([#18954](https://github.com/Lightning-AI/pytorch-lightning/issues/18954))
+- Fixed an issue preventing the user to `Trainer.save_checkpoint()` an FSDP model when `Trainer.test/validate/predict()` ran after `Trainer.fit()` ([#18992](https://github.com/Lightning-AI/pytorch-lightning/issues/18992))
 
 
 ## [2.1.1] - 2023-11-06
 
 ### Fixed
 
 - Fixed an issue when replacing an existing `last.ckpt` file with a symlink ([#18793](https://github.com/Lightning-AI/lightning/pull/18793))
-- Fixed an issue when `BatchSizeFinder` `steps_per_trial` parameter ends up defining how many validation batches to run during the entire training ([#18394](https://github.com/Lightning-AI/lightning/issues/18394))
-- Fixed an issue saving the `last.ckpt` file when using `ModelCheckpoint` on a remote filesystem and no logger is used ([#18867](https://github.com/Lightning-AI/lightning/issues/18867))
+- Fixed an issue when `BatchSizeFinder` `steps_per_trial` parameter ends up defining how many validation batches to run during the entire training ([#18394](https://github.com/Lightning-AI/pytorch-lightning/issues/18394))
+- Fixed an issue saving the `last.ckpt` file when using `ModelCheckpoint` on a remote filesystem and no logger is used ([#18867](https://github.com/Lightning-AI/pytorch-lightning/issues/18867))
 - Refined the FSDP saving logic and error messaging when path exists ([#18884](https://github.com/Lightning-AI/lightning/pull/18884))
-- Fixed an issue parsing the version from folders that don't include a version number in `TensorBoardLogger` and `CSVLogger` ([#18897](https://github.com/Lightning-AI/lightning/issues/18897))
+- Fixed an issue parsing the version from folders that don't include a version number in `TensorBoardLogger` and `CSVLogger` ([#18897](https://github.com/Lightning-AI/pytorch-lightning/issues/18897))
 
 
 ## [2.1.0] - 2023-10-11
diff --git a/src/lightning/pytorch/callbacks/stochastic_weight_avg.py b/src/lightning/pytorch/callbacks/stochastic_weight_avg.py
@@ -354,7 +354,7 @@ def _clear_schedulers(trainer: "pl.Trainer") -> None:
         # Note that this relies on the callback state being restored before the scheduler state is
         # restored, and doesn't work if restore_checkpoint_after_setup is True, but at the time of
         # writing that is only True for deepspeed which is already not supported by SWA.
-        # See https://github.com/Lightning-AI/lightning/issues/11665 for background.
+        # See https://github.com/Lightning-AI/pytorch-lightning/issues/11665 for background.
         if trainer.lr_scheduler_configs:
             assert len(trainer.lr_scheduler_configs) == 1
             trainer.lr_scheduler_configs.clear()
diff --git a/src/lightning/pytorch/plugins/precision/xla.py b/src/lightning/pytorch/plugins/precision/xla.py
@@ -79,7 +79,7 @@ def optimizer_step(  # type: ignore[override]
             # we lack coverage here so disable this - something to explore if there's demand
             raise MisconfigurationException(
                 "Skipping backward by returning `None` from your `training_step` is not implemented with XLA."
-                " Please, open an issue in `https://github.com/Lightning-AI/lightning/issues`"
+                " Please, open an issue in `https://github.com/Lightning-AI/pytorch-lightning/issues`"
                 " requesting this feature."
             )
         return closure_result
diff --git a/src/lightning/pytorch/strategies/launchers/multiprocessing.py b/src/lightning/pytorch/strategies/launchers/multiprocessing.py
@@ -88,7 +88,7 @@ def __init__(
     def is_interactive_compatible(self) -> bool:
         # The start method 'spawn' is not supported in interactive environments
         # The start method 'fork' is the only one supported in Jupyter environments, with constraints around CUDA
-        # initialization. For more context, see https://github.com/Lightning-AI/lightning/issues/7550
+        # initialization. For more context, see https://github.com/Lightning-AI/pytorch-lightning/issues/7550
         return self._start_method == "fork"
 
     @override
@@ -111,7 +111,7 @@ def launch(self, function: Callable, *args: Any, trainer: Optional["pl.Trainer"]
         if self._start_method == "spawn":
             _check_missing_main_guard()
         if self._already_fit and trainer is not None and trainer.state.fn == TrainerFn.FITTING:
-            # resolving https://github.com/Lightning-AI/lightning/issues/18775 will lift this restriction
+            # resolving https://github.com/Lightning-AI/pytorch-lightning/issues/18775 will lift this restriction
             raise NotImplementedError(
                 "Calling `trainer.fit()` twice on the same Trainer instance using a spawn-based strategy is not"
                 " supported. You can work around this limitation by creating a new Trainer instance and passing the"
diff --git a/src/lightning/pytorch/strategies/launchers/xla.py b/src/lightning/pytorch/strategies/launchers/xla.py
@@ -76,7 +76,7 @@ def launch(self, function: Callable, *args: Any, trainer: Optional["pl.Trainer"]
 
         """
         if self._already_fit and trainer is not None and trainer.state.fn == TrainerFn.FITTING:
-            # resolving https://github.com/Lightning-AI/lightning/issues/18775 will lift this restriction
+            # resolving https://github.com/Lightning-AI/pytorch-lightning/issues/18775 will lift this restriction
             raise NotImplementedError(
                 "Calling `trainer.fit()` twice on the same Trainer instance using a spawn-based strategy is not"
                 " supported. You can work around this by creating a new Trainer instance and passing the"
diff --git a/src/lightning/pytorch/trainer/connectors/logger_connector/fx_validator.py b/src/lightning/pytorch/trainer/connectors/logger_connector/fx_validator.py
@@ -154,7 +154,7 @@ def check_logging(cls, fx_name: str) -> None:
         if fx_name not in cls.functions:
             raise RuntimeError(
                 f"Logging inside `{fx_name}` is not implemented."
-                " Please, open an issue in `https://github.com/Lightning-AI/lightning/issues`."
+                " Please, open an issue in `https://github.com/Lightning-AI/pytorch-lightning/issues`."
             )
 
         if cls.functions[fx_name] is None:
diff --git a/src/lightning_fabric/__setup__.py b/src/lightning_fabric/__setup__.py
@@ -85,7 +85,7 @@ def _setup_args() -> dict[str, Any]:
         },
         "extras_require": _prepare_extras(),
         "project_urls": {
-            "Bug Tracker": "https://github.com/Lightning-AI/lightning/issues",
+            "Bug Tracker": "https://github.com/Lightning-AI/pytorch-lightning/issues",
             "Documentation": "https://pytorch-lightning.rtfd.io/en/latest/",
             "Source Code": "https://github.com/Lightning-AI/lightning",
         },
diff --git a/src/pytorch_lightning/__setup__.py b/src/pytorch_lightning/__setup__.py
@@ -89,7 +89,7 @@ def _setup_args() -> dict[str, Any]:
         ),
         "extras_require": _prepare_extras(),
         "project_urls": {
-            "Bug Tracker": "https://github.com/Lightning-AI/lightning/issues",
+            "Bug Tracker": "https://github.com/Lightning-AI/pytorch-lightning/issues",
             "Documentation": "https://pytorch-lightning.rtfd.io/en/latest/",
             "Source Code": "https://github.com/Lightning-AI/lightning",
         },
diff --git a/tests/tests_pytorch/callbacks/test_early_stopping.py b/tests/tests_pytorch/callbacks/test_early_stopping.py
@@ -61,8 +61,8 @@ def on_train_epoch_end(self, trainer, pl_module):
 def test_resume_early_stopping_from_checkpoint(tmp_path):
     """Prevent regressions to bugs:
 
-    https://github.com/Lightning-AI/lightning/issues/1464
-    https://github.com/Lightning-AI/lightning/issues/1463
+    https://github.com/Lightning-AI/pytorch-lightning/issues/1464
+    https://github.com/Lightning-AI/pytorch-lightning/issues/1463
 
     """
     seed_everything(42)
diff --git a/tests/tests_pytorch/callbacks/test_stochastic_weight_avg.py b/tests/tests_pytorch/callbacks/test_stochastic_weight_avg.py
diff --git a/tests/tests_pytorch/core/test_datamodules.py b/tests/tests_pytorch/core/test_datamodules.py
diff --git a/tests/tests_pytorch/profilers/test_profiler.py b/tests/tests_pytorch/profilers/test_profiler.py
diff --git a/tests/tests_pytorch/trainer/logging_/test_train_loop_logging.py b/tests/tests_pytorch/trainer/logging_/test_train_loop_logging.py

Original file line number	Diff line number	Diff line change
`@@ -319,7 +319,7 @@ Additionally, DeepSpeed supports offloading to NVMe drives for even larger model`
`319`	`319`	`)`
`320`	`320`	`trainer.fit(model)`
`321`	`321`
`322`		-When offloading to NVMe you may notice that the speed is slow. There are parameters that need to be tuned based on the drives that you are using. Running the `aio_bench_perf_sweep.py <https://github.com/microsoft/DeepSpeed/blob/master/csrc/aio/py_test/aio_bench_perf_sweep.py>`__ script can help you to find optimum parameters. See the `issue <https://github.com/microsoft/DeepSpeed/issues/998>`__ for more information on how to parse the information.
	`322`	+When offloading to NVMe you may notice that the speed is slow. There are parameters that need to be tuned based on the drives that you are using. Running the `aio_bench_perf_sweep.py <https://github.com/microsoft/DeepSpeed/blob/master/csrc/aio/py_test/aio_bench_perf_sweep.py>`__ script can help you to find optimum parameters. See the `issue <https://github.com/deepspeedai/DeepSpeed/issues/998>`__ for more information on how to parse the information.
`323`	`323`
`324`	`324`	`.. _deepspeed-activation-checkpointing:`
`325`	`325`