Use latest stable PyTorch/DeepSpeed for Push & Scheduled CI (#17417)

* update versions
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

.github/workflows/self-push.yml

@@ -198,19 +198,12 @@ jobs:
         machine_type: [single-gpu]
     runs-on: [self-hosted, docker-gpu, '${{ matrix.machine_type }}']
     container:
-      image: nvcr.io/nvidia/pytorch:21.03-py3
+      image: huggingface/transformers-pytorch-deepspeed-latest-gpu
       options: --gpus 0 --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     steps:
-      - name: Checkout transformers
-        uses: actions/checkout@v2
-        with:
-          fetch-depth: 2
-
-      - name: Install dependencies
-        run: |
-          apt -y update && apt install -y libaio-dev
-          pip install --upgrade pip
-          pip install .[deepspeed-testing]
+      - name: Update clone
+        working-directory: /workspace/transformers
+        run: git fetch && git checkout ${{ github.sha }}
 
       - name: NVIDIA-SMI
         run: |
@@ -247,30 +240,24 @@ jobs:
         machine_type: [multi-gpu]
     runs-on: [self-hosted, docker-gpu, '${{ matrix.machine_type }}']
     container:
-      image: nvcr.io/nvidia/pytorch:21.03-py3
+      image: huggingface/transformers-pytorch-deepspeed-latest-gpu
       options: --gpus all --shm-size "16gb" --ipc host -v /mnt/cache/.cache/huggingface:/mnt/cache/
     steps:
-      - name: Checkout transformers
-        uses: actions/checkout@v2
-        with:
-          fetch-depth: 2
-
-      - name: Install dependencies
-        run: |
-          apt -y update && apt install -y libaio-dev
-          pip install --upgrade pip
-          rm -rf ~/.cache/torch_extensions/ # shared between conflicting builds
-          pip install .[testing,deepspeed,fairscale]
+      - name: Update clone
+        working-directory: /workspace/transformers
+        run: git fetch && git checkout ${{ github.sha }}
 
       - name: NVIDIA-SMI
         run: |
           nvidia-smi
 
       - name: Environment
+        working-directory: /workspace/transformers
         run: |
           python utils/print_env.py
 
       - name: Run all non-slow selected tests on GPU
+        working-directory: /workspace/transformers
         # TODO: Here we pass all tests in the 2 folders for simplicity. It's better to pass only the identified tests.
         run: |
           python -m pytest -n 1 --dist=loadfile -v --make-reports=${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu tests/deepspeed tests/extended
@@ -278,14 +265,14 @@ jobs:
       - name: Failure short reports
         if: ${{ failure() }}
         continue-on-error: true
-        run: cat reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu/failures_short.txt
+        run: cat /workspace/transformers/reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu/failures_short.txt
 
       - name: Test suite reports artifacts
         if: ${{ always() }}
         uses: actions/upload-artifact@v2
         with:
           name: ${{ matrix.machine_type }}_run_tests_torch_cuda_extensions_gpu_test_reports
-          path: reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu
+          path: /workspace/transformers/reports/${{ matrix.machine_type }}_tests_torch_cuda_extensions_gpu
 
   send_results:
     name: Send results to webhook

.github/workflows/self-scheduled.yml

@@ -306,14 +306,6 @@ jobs:
         working-directory: /workspace/transformers
         run: git fetch && git checkout ${{ github.sha }}
 
-      - name: Re-compile DeepSpeed
-        working-directory: /workspace
-        run: |
-          pip install deepspeed # installs the deps correctly
-          rm -rf DeepSpeed
-          git clone https://github.com/microsoft/DeepSpeed && cd DeepSpeed && rm -rf build
-          DS_BUILD_CPU_ADAM=1 DS_BUILD_AIO=1 DS_BUILD_UTILS=1 python3 -m pip install -e . --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check
-
       - name: NVIDIA-SMI
         run: |
           nvidia-smi

docker/transformers-all-latest-gpu/Dockerfile

@@ -11,9 +11,13 @@ ARG REF=main
 RUN git clone https://github.com/huggingface/transformers && cd transformers && git checkout $REF
 RUN python3 -m pip install --no-cache-dir -e ./transformers[dev,onnxruntime]
 
-RUN python3 -m pip install --no-cache-dir -U torch tensorflow
+RUN python3 -m pip install --no-cache-dir -U torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu113
+RUN python3 -m pip install --no-cache-dir -U tensorflow
 RUN python3 -m pip uninstall -y flax jax
-RUN python3 -m pip install --no-cache-dir torch-scatter -f https://data.pyg.org/whl/torch-$(python3 -c "from torch import version; print(version.__version__.split('+')[0])")+cu102.html
+
+RUN python3 -m pip install --no-cache-dir torch-scatter -f https://data.pyg.org/whl/torch-$(python3 -c "from torch import version; print(version.__version__.split('+')[0])")+cu113.html
+RUN python3 -m pip install --no-cache-dir intel_extension_for_pytorch==$(python3 -c "from torch import version; print(version.__version__.split('+')[0])")+cpu -f https://software.intel.com/ipex-whl-stable
+
 RUN python3 -m pip install --no-cache-dir git+https://github.com/facebookresearch/detectron2.git pytesseract https://github.com/kpu/kenlm/archive/master.zip
 RUN python3 -m pip install -U "itsdangerous<2.1.0"
 

docker/transformers-pytorch-deepspeed-latest-gpu/Dockerfile

@@ -11,8 +11,8 @@ ARG REF=main
 RUN git clone https://github.com/huggingface/transformers && cd transformers && git checkout $REF
 RUN python3 -m pip install --no-cache-dir -e ./transformers[deepspeed-testing]
 
-RUN git clone https://github.com/microsoft/DeepSpeed && cd DeepSpeed && rm -rf build && \
-    DS_BUILD_CPU_ADAM=1 DS_BUILD_AIO=1 DS_BUILD_UTILS=1 python3 -m pip install -e . --global-option="build_ext" --global-option="-j8" --no-cache -v --disable-pip-version-check 2>&1
+# Install latest release PyTorch
+RUN python3 -m pip install --no-cache-dir -U torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu113
 
 # When installing in editable mode, `transformers` is not recognized as a package.
 # this line must be added in order for python to be aware of transformers.

docker/transformers-pytorch-gpu/Dockerfile

@@ -13,11 +13,16 @@ RUN python3 -m pip install --no-cache-dir -e ./transformers[dev-torch,testing]
 
 # If set to nothing, will install the latest version
 ARG PYTORCH=''
+ARG TORCH_VISION=''
+ARG TORCH_AUDIO=''
+
+RUN [ ${#PYTORCH} -gt 0 ] && VERSION='torch=='$PYTORCH'.*' ||  VERSION='torch'; python3 -m pip install --no-cache-dir -U $VERSION --extra-index-url https://download.pytorch.org/whl/cu113
+RUN [ ${#TORCH_VISION} -gt 0 ] && VERSION='torchvision=='TORCH_VISION'.*' ||  VERSION='torchvision'; python3 -m pip install --no-cache-dir -U $VERSION --extra-index-url https://download.pytorch.org/whl/cu113
+RUN [ ${#TORCH_AUDIO} -gt 0 ] && VERSION='torchaudio=='TORCH_AUDIO'.*' ||  VERSION='torchaudio'; python3 -m pip install --no-cache-dir -U $VERSION --extra-index-url https://download.pytorch.org/whl/cu113
 
-RUN [ ${#PYTORCH} -gt 0 ] && VERSION='torch=='$PYTORCH'.*' ||  VERSION='torch'; python3 -m pip install --no-cache-dir -U $VERSION
 RUN python3 -m pip uninstall -y tensorflow flax
 
-RUN python3 -m pip install --no-cache-dir torch-scatter -f https://data.pyg.org/whl/torch-$(python3 -c "from torch import version; print(version.__version__.split('+')[0])")+cu102.html
+RUN python3 -m pip install --no-cache-dir torch-scatter -f https://data.pyg.org/whl/torch-$(python3 -c "from torch import version; print(version.__version__.split('+')[0])")+cu113.html
 RUN python3 -m pip install --no-cache-dir git+https://github.com/facebookresearch/detectron2.git pytesseract https://github.com/kpu/kenlm/archive/master.zip
 RUN python3 -m pip install -U "itsdangerous<2.1.0"