Merge branch 'master' into ci/tpu

Lightning-AI · Dec 2, 2022 · 8644a16 · 8644a16
2 parents a771df4 + fee52f9
commit 8644a16
Show file tree

Hide file tree

Showing 33 changed files with 283 additions and 199 deletions.
diff --git a/.azure/gpu-benchmark.yml b/.azure/gpu-benchmark.yml
@@ -68,14 +68,7 @@ jobs:
         pip list
       displayName: 'Image info & NVIDIA'
 
-    - bash: |
-        python .actions/assistant.py requirements_prune_pkgs --packages [horovod,bagua,colossalai] --req_files [requirements/pytorch/strategies.txt]
-
-        PYTORCH_VERSION=$(python -c "import torch; print(torch.__version__.split('+')[0])")
-        python ./requirements/pytorch/adjust-versions.py requirements/pytorch/base.txt ${PYTORCH_VERSION}
-      displayName: 'Adjust dependencies'
-
-    - bash: pip install -e .[dev,strategies,examples] --find-links ${TORCH_URL}
+    - bash: pip install -e .[dev] --find-links ${TORCH_URL}
       env:
         PACKAGE_NAME: "pytorch"
         FREEZE_REQUIREMENTS: "1"

diff --git a/.github/actions/pkg-publish/action.yml b/.github/actions/pkg-publish/action.yml
@@ -5,9 +5,14 @@ inputs:
   pkg-pattern:
     description: what file pattern is searched in folder, so for example `*_app*`
     required: true
+  pypi-test-token:
+    description: login token for PyPI
+    default: ''
+    required: false
   pypi-token:
     description: login token for PyPI
-    required: true
+    default: ''
+    required: false
 
 runs:
   using: "composite"
@@ -20,16 +25,19 @@ runs:
     shell: bash
 
   # We do this, since failures on test.pypi aren't that bad
-  #- name: Publish to Test PyPI
-  #  uses: pypa/gh-action-pypi-publish@v1.5.1
-  #  with:
-  #    user: __token__
-  #    password: ${{ secrets.test_pypi_token_lai }}
-  #    repository_url: https://test.pypi.org/legacy/
-  #    verbose: true
+  - name: Publish to Test PyPI
+    uses: pypa/gh-action-pypi-publish@v1.5.1
+    if: inputs.pypi-test-token != ''
+    with:
+      user: __token__
+      password: ${{ secrets.test_pypi_token_lai }}
+      repository_url: https://test.pypi.org/legacy/
+      packages_dir: pypi/
+      verbose: true
 
   - name: Publish distribution 📦 to PyPI
     uses: pypa/gh-action-pypi-publish@v1.5.1
+    if: inputs.pypi-token != ''
     with:
       user: __token__
       password: ${{ inputs.pypi-token }}

diff --git a/.github/workflows/legacy-checkpoints.yml b/.github/workflows/legacy-checkpoints.yml
@@ -41,11 +41,15 @@ on:
       AWS_SECRET_KEY_ID:
         required: true
 
+defaults:
+  run:
+    shell: bash
+
 jobs:
   create-legacy-ckpts:
     runs-on: ubuntu-20.04
     outputs:
-      pl-version: ${{ steps.decide-pl-version.outputs.pl-version }}
+      pl-version: ${{ steps.decide-version.outputs.pl-version }}
     defaults:
       run:
         working-directory: tests/legacy
@@ -69,16 +73,24 @@ jobs:
       env:
         PACKAGE_NAME: pytorch
       run: |
-        pip install -e . -f https://download.pytorch.org/whl/cpu/torch_stable.html
+        pip install . -f https://download.pytorch.org/whl/cpu/torch_stable.html
         pip list
       if: inputs.pl_version == ''
 
+    - name: Install PL version
+      run: |
+        pip install "pytorch-lightning==${{ inputs.pl_version }}" \
+          -f https://download.pytorch.org/whl/cpu/torch_stable.html \
+          --extra-index-url https://test.pypi.org/simple/
+        pip list
+      if: inputs.pl_version != ''
+
     - name: Pull legacy checkpoints
       working-directory: ./
       run: bash .actions/pull_legacy_checkpoints.sh
 
     - name: Decide PL version to create a PR with
-      id: decide-pl-version
+      id: decide-version
       run: |
         python -c "import pytorch_lightning as pl; print(f'pl-version={pl.__version__}')" >> $GITHUB_OUTPUT || echo pl-version='' >> $GITHUB_OUTPUT
 

diff --git a/.github/workflows/release-pypi.yml b/.github/workflows/release-pypi.yml
@@ -7,6 +7,10 @@ on:
   release:
     types: [published]
 
+defaults:
+  run:
+    shell: bash
+
 jobs:
   init:
     runs-on: ubuntu-20.04
@@ -24,8 +28,7 @@ jobs:
     runs-on: ubuntu-20.04
     strategy:
       fail-fast: true
-      # run sequential
-      max-parallel: 1
+      max-parallel: 1  # run sequential to prevent download/upload collisions
       matrix:
         pkg-name: ["lightning", "app", "lite", "pytorch"]
     steps:
@@ -70,7 +73,69 @@ jobs:
         files: 'dist/*'
         repo-token: ${{ secrets.GITHUB_TOKEN }}
 
-  publish-packages:
+  release-version:
+    runs-on: ubuntu-20.04
+    outputs:
+      tag: ${{ steps.lai-package.outputs.version }}
+    steps:
+    - uses: actions/checkout@v3
+    - name: install Package
+      env:
+        PACKAGE_NAME: "lightning"
+      run: pip install . -f https://download.pytorch.org/whl/cpu/torch_stable.html
+    - name: package Version
+      id: lai-package
+      run: python -c "import lightning as L; print(f'version={L.__version__}')" >> $GITHUB_OUTPUT
+
+  signaling:
+    runs-on: ubuntu-20.04
+    needs: [release-version, pre-publish-packages]
+    env:
+      TAG: ${{ needs.release-version.outputs.tag }}
+    steps:
+    - uses: actions/setup-python@v4
+      with:
+        python-version: 3.8
+    - uses: actions/checkout@v3
+      with:
+        repository: gridai/base-images
+        token: ${{ secrets.PAT_GHOST }}
+        ref: main
+    - uses: fregante/setup-git-token@v1
+      with:
+        token: ${{ secrets.PAT_GHOST }}
+        name: PL Ghost
+        email: pl-github@grid.ai
+    - name: Update lightning version
+      run: |
+        import json, os
+        with open("versions.json") as fo:
+            vers = json.load(fo)
+        vers["lightning_version"] = os.getenv('TAG')
+        with open("versions.json", "w") as fw:
+            json.dump(vers, fw)
+      shell: python
+    - name: GIT Commit
+      run: |
+        git add versions.json
+        git commit -m "bumping lightning version -> ${TAG}"
+        cat versions.json
+    - name: GIT Push
+      run: |
+        git status
+        git push
+
+  waiting:
+    # TODO: replace with back signal from build images/ loop checking for a specific branch?
+    runs-on: ubuntu-20.04
+    needs: signaling
+    steps:
+    - name: Delay releasing
+      uses: juliangruber/sleep-action@v1
+      with:
+        time: 30m
+
+  pre-publish-packages:
     runs-on: ubuntu-20.04
     needs: build-packages
     if: startsWith(github.event.ref, 'refs/tags') || github.event_name == 'release'
@@ -82,38 +147,59 @@ jobs:
         path: dist
     - run: ls -lh dist/
     - run: mkdir pypi/
-
-    - name: Delay releasing
-      uses: juliangruber/sleep-action@v1
+    - uses: ./.github/actions/pkg-publish
       with:
-        time: 10m
+        pkg-pattern: "*app*"
+        pypi-test-token: ${{ secrets.PYPI_TEST_TOKEN_APP }}
+    - uses: ./.github/actions/pkg-publish
+      with:
+        pkg-pattern: "*lite*"
+        pypi-test-token: ${{ secrets.PYPI_TEST_TOKEN_LITE }}
+    - uses: ./.github/actions/pkg-publish
+      with:
+        pkg-pattern: "*pytorch*"
+        pypi-test-token: ${{ secrets.PYPI_TEST_TOKEN_PYTORCH }}
+    - uses: ./.github/actions/pkg-publish
+      with:
+        pkg-pattern: "*"
+        pypi-test-token: ${{ secrets.PYPI_TEST_TOKEN_LAI }}
 
+  publish-packages:
+    runs-on: ubuntu-20.04
+    needs: waiting
+    if: startsWith(github.event.ref, 'refs/tags') || github.event_name == 'release'
+    steps:
+    - uses: actions/checkout@v3
+    - uses: actions/download-artifact@v3
+      with:
+        name: dist-packages-${{ github.sha }}
+        path: dist
+    - run: ls -lh dist/
+    - run: mkdir pypi/
     - uses: ./.github/actions/pkg-publish
       with:
         pkg-pattern: "*app*"
         pypi-token: ${{ secrets.PYPI_TOKEN_APP }}
-
     - uses: ./.github/actions/pkg-publish
       with:
         pkg-pattern: "*lite*"
         pypi-token: ${{ secrets.PYPI_TOKEN_LITE }}
-
     - uses: ./.github/actions/pkg-publish
       with:
         pkg-pattern: "*pytorch*"
         pypi-token: ${{ secrets.PYPI_TOKEN_PYTORCH }}
-
     - uses: ./.github/actions/pkg-publish
       with:
         pkg-pattern: "*"
         pypi-token: ${{ secrets.PYPI_TOKEN_LAI }}
 
   legacy-checkpoints:
-    needs: publish-packages
+    needs: [release-version, pre-publish-packages]
     uses: ./.github/workflows/legacy-checkpoints.yml
     with:
       push_to_s3: true
       create_pr: true
+      pl_version: ${{ needs.release-version.outputs.tag }}
     secrets:
       AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_ID }}
       AWS_SECRET_KEY_ID: ${{ secrets.AWS_SECRET_KEY_ID }}
diff --git a/docs/source-pytorch/advanced/model_parallel.rst b/docs/source-pytorch/advanced/model_parallel.rst
@@ -1108,9 +1108,6 @@ Combine hooks for accumulated benefit:
 
 When using Post-localSGD, you must also pass ``model_averaging_period`` to allow for model parameter averaging:
 
-.. note::
-    Post-localSGD support requires PyTorch>=1.10.0
-
 .. code-block:: python
 
     from pytorch_lightning import Trainer

diff --git a/docs/source-pytorch/common/precision_intermediate.rst b/docs/source-pytorch/common/precision_intermediate.rst
@@ -115,8 +115,6 @@ BFloat16 Mixed Precision
 
 .. warning::
 
-    BFloat16 requires PyTorch 1.10 or later and is only supported with PyTorch Native AMP.
-
     BFloat16 is also experimental and may not provide significant speedups or memory improvements, offering better numerical stability.
 
     Do note for GPUs, the most significant benefits require `Ampere <https://en.wikipedia.org/wiki/Ampere_(microarchitecture)>`__ based GPUs, such as A100s or 3090s.
@@ -126,14 +124,13 @@ BFloat16 Mixed precision is similar to FP16 mixed precision, however, it maintai
 Under the hood, we use `torch.autocast <https://pytorch.org/docs/stable/amp.html>`__ with the dtype set to ``bfloat16``, with no gradient scaling.
 
 .. testcode::
-    :skipif: not _TORCH_GREATER_EQUAL_1_10 or not torch.cuda.is_available()
+    :skipif: not torch.cuda.is_available()
 
     Trainer(accelerator="gpu", devices=1, precision="bf16")
 
 It is also possible to use BFloat16 mixed precision on the CPU, relying on MKLDNN under the hood.
 
 .. testcode::
-    :skipif: not _TORCH_GREATER_EQUAL_1_10
 
     Trainer(precision="bf16")
 

diff --git a/docs/source-pytorch/conf.py b/docs/source-pytorch/conf.py
@@ -400,7 +400,6 @@ def package_list_from_file(file):
 from pytorch_lightning.utilities import (
     _APEX_AVAILABLE,
     _TORCHVISION_AVAILABLE,
-    _TORCH_GREATER_EQUAL_1_10,
 )
 from pytorch_lightning.loggers.neptune import _NEPTUNE_AVAILABLE
 from pytorch_lightning.loggers.comet import _COMET_AVAILABLE

diff --git a/requirements/app/base.txt b/requirements/app/base.txt
@@ -12,4 +12,3 @@ beautifulsoup4>=4.8.0, <4.11.2
 inquirer>=2.10.0
 psutil<5.9.4
 click<=8.1.3
-lightning_api_access>=0.0.3
diff --git a/requirements/app/components.txt b/requirements/app/components.txt
@@ -0,0 +1,2 @@
+# deps required by components in the lightning app repository (src/lightning_app/components)
+lightning_api_access>=0.0.3
diff --git a/requirements/app/devel.txt b/requirements/app/devel.txt
@@ -10,4 +10,7 @@
 # extended list of dependencies for UI
 -r ./ui.txt
 
+# extended list of dependencies for UI
+-r ./components.txt
+
 -r ./examples.txt
diff --git a/src/lightning_app/CHANGELOG.md b/src/lightning_app/CHANGELOG.md
@@ -42,6 +42,8 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 - Fixed the `enable_spawn` method of the `WorkRunExecutor` ([#15812](https://github.com/Lightning-AI/lightning/pull/15812)
 
+- Fixed Sigterm Handler causing thread lock which caused KeyboardInterrupt to hang ([#15881](https://github.com/Lightning-AI/lightning/pull/15881))
+
 
 ## [1.8.3] - 2022-11-22
 
@@ -94,6 +96,8 @@ The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/).
 
 ## [1.8.1] - 2022-11-10
 
+- Fixed bug when launching apps on multiple clusters ([#15484](https://github.com/Lightning-AI/lightning/pull/15484))
+
 
 ### Added
 

diff --git a/src/lightning_app/components/serve/python_server.py b/src/lightning_app/components/serve/python_server.py
@@ -6,6 +6,7 @@
 
 import uvicorn
 from fastapi import FastAPI
+from lightning_utilities.core.imports import module_available
 from pydantic import BaseModel
 from starlette.staticfiles import StaticFiles
 
@@ -17,9 +18,14 @@
 
 logger = Logger(__name__)
 
+__doctest_skip__ = []
+# Skip doctests if requirements aren't available
+if not module_available("lightning_api_access"):
+    __doctest_skip__ += ["PythonServer", "PythonServer.*"]
+
 # Skip doctests if requirements aren't available
 if not _is_torch_available():
-    __doctest_skip__ = ["PythonServer", "PythonServer.*"]
+    __doctest_skip__ += ["PythonServer", "PythonServer.*"]
 
 
 class _PyTorchSpawnRunExecutor(WorkRunExecutor):
@@ -90,7 +96,7 @@ def _get_sample_data() -> Dict[Any, Any]:
 
 
 class PythonServer(LightningWork, abc.ABC):
-    @requires("torch")
+    @requires(["torch", "lightning_api_access"])
     def __init__(  # type: ignore
         self,
         host: str = "127.0.0.1",
Original file line number	Diff line number	Diff line change
		@@ -0,0 +1,2 @@
		# deps required by components in the lightning app repository (src/lightning_app/components)
		lightning_api_access>=0.0.3