ci/gpu: fix install future & use local cache (#16929)

.azure/gpu-tests-fabric.yml

@@ -47,11 +47,12 @@ jobs:
     variables:
       DEVICES: $( python -c 'print("$(Agent.Name)".split("_")[-1])' )
       FREEZE_REQUIREMENTS: "1"
+      PIP_CACHE_DIR: "/var/tmp/pip"
     container:
       image: "pytorchlightning/pytorch_lightning:base-cuda-py3.9-torch1.13-cuda11.7.1"
       # default shm size is 64m. Increase it to avoid:
       # 'Error while creating shared memory: unhandled system error, NCCL version 2.7.8'
-      options: "--gpus=all --shm-size=2gb"
+      options: "--gpus=all --shm-size=2gb  -v /var/tmp:/var/tmp"
     strategy:
       matrix:
         'pkg: Fabric':

.azure/gpu-tests-pytorch.yml

@@ -56,7 +56,7 @@ jobs:
           scope: ""
           PACKAGE_NAME: "pytorch"
         'PyTorch | future':
-          image: "pytorchlightning/pytorch_lightning:base-cuda-py3.9-torch1.13-cuda11.7.1"
+          image: "pytorchlightning/pytorch_lightning:base-cuda-py3.10-torch1.13-cuda11.7.1"
           scope: "future"
           PACKAGE_NAME: "pytorch"
         'Lightning | latest':
@@ -67,11 +67,12 @@ jobs:
     variables:
       DEVICES: $( python -c 'print("$(Agent.Name)".split("_")[-1])' )
       FREEZE_REQUIREMENTS: "1"
+      PIP_CACHE_DIR: "/var/tmp/pip"
     container:
       image: $(image)
       # default shm size is 64m. Increase it to avoid:
       # 'Error while creating shared memory: unhandled system error, NCCL version 2.7.8'
-      options: "--gpus=all --shm-size=2gb"
+      options: "--gpus=all --shm-size=2gb  -v /var/tmp:/var/tmp"
     workspace:
       clean: all
     steps:
@@ -114,8 +115,11 @@ jobs:
       displayName: 'Install package & extras'
 
     - bash: |
+        pip install -U -q pip
         pip uninstall -y torch torchvision
-        pip install torch torchvision -U --pre --no-cache -f https://download.pytorch.org/whl/test/cu${CUDA_VERSION_MM}/torch_test.html
+        pip install "torch==2.0.0" "torchvision==0.15.0" -U --pre \
+          -f "https://download.pytorch.org/whl/test/cu${CUDA_VERSION_MM}/torch_test.html" \
+          -f "https://download.pytorch.org/whl/nightly/cu${CUDA_VERSION_MM}/torch_nightly.html"
         python -c "from torch import __version__ as ver; assert ver.startswith('2.0.0'), ver"
       condition: eq(variables['scope'], 'future')
       displayName: 'bump to future'

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

@@ -8,7 +8,7 @@ on:
     types: [opened, reopened, ready_for_review, synchronize]  # added `ready_for_review` since draft is skipped
     paths:
       - ".actions/**"
-      - ".github/workflows/ci-pytorch-dockers.yml"
+      - ".github/workflows/ci-dockers-pytorch.yml"
       - "dockers/**"
       - "requirements/*.txt"
       - "requirements/pytorch/**"
@@ -105,6 +105,7 @@ jobs:
           - {python_version: "3.9", pytorch_version: "1.11", cuda_version: "11.3.1"}
           - {python_version: "3.9", pytorch_version: "1.12", cuda_version: "11.6.1"}
           - {python_version: "3.9", pytorch_version: "1.13", cuda_version: "11.7.1"}
+          - {python_version: "3.10", pytorch_version: "1.13", cuda_version: "11.7.1"}
     steps:
       - uses: actions/checkout@v3
       - uses: docker/setup-buildx-action@v2