Merge branch 'master' into dependabot-pip-requirements-torch-2.9.0

Author: deependujha

.azure/gpu-tests-fabric.yml

@@ -76,7 +76,7 @@ jobs:
           cuda_ver=$(python -c "import torch ; print(''.join(map(str, torch.version.cuda.split('.')[:2])))")
           echo "##vso[task.setvariable variable=CUDA_VERSION_MM]$cuda_ver"
           echo "##vso[task.setvariable variable=TORCH_URL]https://download.pytorch.org/whl/cu${cuda_ver}/torch_stable.html"
-          scope=$(python -c 'n = "$(PACKAGE_NAME)" ; print(dict(fabric="lightning_fabric").get(n, n))')
+          scope=$(python -c 'n = "${PACKAGE_NAME}" ; print(dict(fabric="lightning_fabric").get(n, n))')
           echo "##vso[task.setvariable variable=COVERAGE_SOURCE]$scope"
         displayName: "set env. vars"
       - bash: |
@@ -140,7 +140,7 @@ jobs:
 
       - bash: |
           set -e
-          extra=$(python -c "print({'lightning': 'fabric-'}.get('$(PACKAGE_NAME)', ''))")
+          extra=$(python -c "print({'lightning': 'fabric-'}.get('${PACKAGE_NAME}', ''))")
           pip install -e ".[${extra}dev]" -U --upgrade-strategy=eager --extra-index-url="${TORCH_URL}"
         displayName: "Install package & dependencies"
 

.azure/gpu-tests-pytorch.yml

@@ -75,7 +75,7 @@ jobs:
           cuda_ver=$(python -c "import torch ; print(''.join(map(str, torch.version.cuda.split('.')[:2])))")
           echo "##vso[task.setvariable variable=CUDA_VERSION_MM]$cuda_ver"
           echo "##vso[task.setvariable variable=TORCH_URL]https://download.pytorch.org/whl/cu${cuda_ver}/torch_stable.html"
-          scope=$(python -c 'n = "$(PACKAGE_NAME)" ; print(dict(pytorch="pytorch_lightning").get(n, n))')
+          scope=$(python -c 'n = "${PACKAGE_NAME}" ; print(dict(pytorch="pytorch_lightning").get(n, n))')
           echo "##vso[task.setvariable variable=COVERAGE_SOURCE]$scope"
         displayName: "set env. vars"
       - bash: |
@@ -139,7 +139,7 @@ jobs:
 
       - bash: |
           set -e
-          extra=$(python -c "print({'lightning': 'pytorch-'}.get('$(PACKAGE_NAME)', ''))")
+          extra=$(python -c "print({'lightning': 'pytorch-'}.get('${PACKAGE_NAME}', ''))")
           pip install -e ".[${extra}dev]" -U --upgrade-strategy=eager --extra-index-url="${TORCH_URL}"
         displayName: "Install package & dependencies"
 

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

@@ -133,7 +133,6 @@ jobs:
         run: |
           uv pip install ".[${EXTRA_PREFIX}extra,${EXTRA_PREFIX}test,${EXTRA_PREFIX}strategies]" \
             --upgrade \
-            -r requirements/_integrations/accelerators.txt \
             --find-links="${TORCH_URL}" \
             --find-links="https://download.pytorch.org/whl/torch-tensorrt"
           uv pip list

.lightning/workflows/fabric.yml

@@ -85,7 +85,7 @@ run: |
   CUDA_VERSION_M_M="${CUDA_VERSION%.*}"  # "12.6"
   CUDA_VERSION_MM="${CUDA_VERSION_M_M//./}"  # "126"
   export UV_TORCH_BACKEND=cu${CUDA_VERSION_MM}
-  COVERAGE_SOURCE=$(python -c 'n = "$(PACKAGE_NAME)" ; print(dict(fabric="lightning_fabric").get(n, n))')
+  COVERAGE_SOURCE=$(python -c 'n = "${PACKAGE_NAME}" ; print(dict(fabric="lightning_fabric").get(n, n))')
   echo "collecting coverage for: ${COVERAGE_SOURCE}"
 
   uv pip install fire wget packaging "lightning-utilities[cli]"
@@ -120,7 +120,7 @@ run: |
   fi
 
   echo "Install package with [${PACKAGE_NAME}] extras"
-  extra=$(python -c "print({'lightning': 'fabric-'}.get('$(PACKAGE_NAME)', ''))")
+  extra=$(python -c "print({'lightning': 'fabric-'}.get('${PACKAGE_NAME}', ''))")
   uv pip install ".[${extra}dev]" --upgrade
 
   python requirements/collect_env_details.py

.lightning/workflows/pytorch.yml

@@ -27,7 +27,6 @@ env:
   DEBIAN_FRONTEND: "noninteractive"
   CUDA_TOOLKIT_ROOT_DIR: "/usr/local/cuda"
   MKL_THREADING_LAYER: "GNU"
-  CUDA_LAUNCH_BLOCKING: "1"
   NCCL_DEBUG: "INFO"
   TORCHDYNAMO_VERBOSE: "1"
   FREEZE_REQUIREMENTS: "1"
@@ -50,7 +49,8 @@ run: |
         openmpi-bin \
         ninja-build \
         libnccl2 \
-        libnccl-dev
+        libnccl-dev \
+        unzip
 
   echo "Install Python ${python_version} and UV"
   apt-get install -y python${python_version} python${python_version}-venv python${python_version}-dev
@@ -85,7 +85,7 @@ run: |
   CUDA_VERSION_M_M="${CUDA_VERSION%.*}"  # "12.6"
   CUDA_VERSION_MM="${CUDA_VERSION_M_M//./}"  # "126"
   export UV_TORCH_BACKEND=cu${CUDA_VERSION_MM}
-  COVERAGE_SOURCE=$(python -c 'n = "$(PACKAGE_NAME)" ; print(dict(fabric="pytorch_lightning").get(n, n))')
+  COVERAGE_SOURCE=$(python -c 'n = "${PACKAGE_NAME}" ; print(dict(fabric="pytorch_lightning").get(n, n))')
   echo "collecting coverage for: ${COVERAGE_SOURCE}"
 
   uv pip install -q fire wget packaging "lightning-utilities[cli]"
@@ -120,7 +120,7 @@ run: |
   fi
 
   echo "Install package"
-  extra=$(python -c "print({'lightning': 'pytorch-'}.get('$(PACKAGE_NAME)', ''))")
+  extra=$(python -c "print({'lightning': 'pytorch-'}.get('${PACKAGE_NAME}', ''))")
   uv pip install -e ".[${extra}dev]" --upgrade
 
   echo "Ensure only a single package is installed"

.pre-commit-config.yaml

@@ -27,8 +27,7 @@ repos:
     hooks:
       - id: end-of-file-fixer
       - id: trailing-whitespace
-        # keep formatting in README flexible
-        exclude: README.md
+        exclude: README.md # keep formatting in README flexible
       - id: check-json
       - id: check-yaml
       - id: check-toml

README.md

@@ -64,9 +64,10 @@ Over 340,000 developers use [Lightning Cloud](https://lightning.ai/?utm_source=p
 - [Notebooks](https://lightning.ai/notebooks?utm_source=ptl_readme&utm_medium=referral&utm_campaign=ptl_readme): Persistent GPU workspaces where AI helps you code and analyze.
 - [Inference](https://lightning.ai/deploy?utm_source=ptl_readme&utm_medium=referral&utm_campaign=ptl_readme): Deploy models as inference APIs.   
 
+<a id="why-pytorch-lightning"></a>
 # Why PyTorch Lightning?   
 
-Training models in plain PyTorch is tedious and error-prone - you have to manually handle things like backprop, mixed precision, multi-GPU, and distributed training, often rewriting code for every new project. PyTorch Lightning organizes PyTorch code to automate those complexities so you can focus on your model and data, while keeping full control and scaling from CPU to multi-node without changing your core code. But if you want control of those things, you can still opt into [expert-level control](#lightning-fabric-expert-contro).   
+Training models in plain PyTorch is tedious and error-prone - you have to manually handle things like backprop, mixed precision, multi-GPU, and distributed training, often rewriting code for every new project. PyTorch Lightning organizes PyTorch code to automate those complexities so you can focus on your model and data, while keeping full control and scaling from CPU to multi-node without changing your core code. But if you want control of those things, you can still opt into [expert-level control](#lightning-fabric-expert-control).   
 
 Fun analogy: If PyTorch is Javascript, PyTorch Lightning is ReactJS or NextJS.
 
@@ -215,7 +216,7 @@ Explore various types of training possible with PyTorch Lightning. Pretrain and
 
 | Task | Description | Run |
 |------|--------------|-----|
-| [Hello world](#hello-simple-model) | Pretrain - Hello world example | <a target="_blank" href="https://lightning.ai/lightning-ai/studios/pytorch-lightning-hello-world?utm_source=ptl_readme&utm_medium=referral&utm_campaign=ptl_readme"><img src="https://pl-bolts-doc-images.s3.us-east-2.amazonaws.com/app-2/studio-badge.svg" alt="Open In Studio"/></a> |
+| [Hello world](https://lightning.ai/lightning-ai/studios/pytorch-lightning-hello-world?utm_source=ptl_readme&utm_medium=referral&utm_campaign=ptl_readme) | Pretrain - Hello world example | <a target="_blank" href="https://lightning.ai/lightning-ai/studios/pytorch-lightning-hello-world?utm_source=ptl_readme&utm_medium=referral&utm_campaign=ptl_readme"><img src="https://pl-bolts-doc-images.s3.us-east-2.amazonaws.com/app-2/studio-badge.svg" alt="Open In Studio"/></a> |
 | [Image classification](https://lightning.ai/lightning-ai/studios/image-classification-with-pytorch-lightning?utm_source=ptl_readme&utm_medium=referral&utm_campaign=ptl_readme) | Finetune - ResNet-34 model to classify images of cars | <a target="_blank" href="https://lightning.ai/lightning-ai/studios/image-classification-with-pytorch-lightning?utm_source=ptl_readme&utm_medium=referral&utm_campaign=ptl_readme"><img src="https://pl-bolts-doc-images.s3.us-east-2.amazonaws.com/app-2/studio-badge.svg" alt="Open In Studio"/></a> |
 | [Image segmentation](https://lightning.ai/lightning-ai/studios/image-segmentation-with-pytorch-lightning?utm_source=ptl_readme&utm_medium=referral&utm_campaign=ptl_readme) | Finetune - ResNet-50 model to segment images | <a target="_blank" href="https://lightning.ai/lightning-ai/studios/image-segmentation-with-pytorch-lightning?utm_source=ptl_readme&utm_medium=referral&utm_campaign=ptl_readme"><img src="https://pl-bolts-doc-images.s3.us-east-2.amazonaws.com/app-2/studio-badge.svg" alt="Open In Studio"/></a> |
 | [Object detection](https://lightning.ai/lightning-ai/studios/object-detection-with-pytorch-lightning?utm_source=ptl_readme&utm_medium=referral&utm_campaign=ptl_readme) | Finetune - Faster R-CNN model to detect objects | <a target="_blank" href="https://lightning.ai/lightning-ai/studios/object-detection-with-pytorch-lightning?utm_source=ptl_readme&utm_medium=referral&utm_campaign=ptl_readme"><img src="https://pl-bolts-doc-images.s3.us-east-2.amazonaws.com/app-2/studio-badge.svg" alt="Open In Studio"/></a> |

docs/source-pytorch/common/index.rst

@@ -16,7 +16,6 @@
    Save memory with half-precision <precision>
    ../advanced/model_parallel
    Train on single or multiple GPUs <../accelerators/gpu>
-   Train on single or multiple HPUs <../integrations/hpu/index>
    Train on single or multiple TPUs <../accelerators/tpu>
    Train on MPS <../accelerators/mps>
    Use a pretrained model <../advanced/pretrained>
@@ -161,13 +160,6 @@ How-to Guides
     :col_css: col-md-4
     :height: 180
 
-.. displayitem::
-    :header: Train on single or multiple HPUs
-    :description: Train models faster with HPU accelerators
-    :button_link: ../integrations/hpu/index.html
-    :col_css: col-md-4
-    :height: 180
-
 .. displayitem::
     :header: Train on single or multiple TPUs
     :description: TTrain models faster with TPU accelerators

docs/source-pytorch/common/trainer.rst

@@ -165,6 +165,11 @@ By setting ``workers=True`` in :func:`~lightning.pytorch.seed_everything`, Light
 unique seeds across all dataloader workers and processes for :mod:`torch`, :mod:`numpy` and stdlib
 :mod:`random` number generators. When turned on, it ensures that e.g. data augmentations are not repeated across workers.
 
+.. note::
+
+    If your project depends on NumPy for randomness (e.g. for data
+    augmentation), it is recommended to use version 2.5.0 or higher.
+
 -------
 
 .. _trainer_flags:

docs/source-pytorch/common_usecases.rst

@@ -126,13 +126,6 @@ Customize and extend Lightning for things like custom hardware or distributed st
    :button_link: accelerators/gpu.html
    :height: 100
 
-.. displayitem::
-   :header: Train on single or multiple HPUs
-   :description: Train models faster with HPUs.
-   :col_css: col-md-12
-   :button_link: integrations/hpu/index.html
-   :height: 100
-
 .. displayitem::
    :header: Train on single or multiple TPUs
    :description: Train models faster with TPUs.