Run NVTabular Single GPU tests for CUDA 11 and 12 on RAPIDS Runner

.github/workflows/gpu-tests.yml

@@ -3,18 +3,19 @@ name: GPU Tests
 on:
   workflow_dispatch:
   push:
-    branches: [main]
+    branches:
+      - main
+      - pull-request/*
     tags:
       - "v[0-9]+.[0-9]+.[0-9]+"
-  pull_request:
-    branches: [main]
-    types: [opened, synchronize, reopened, closed]
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
   cancel-in-progress: true
 
 jobs:
+  # Multi-GPU tests
+
   gpu-tests:
     runs-on: 2GPU
 
@@ -31,4 +32,94 @@ jobs:
             raw=$(git branch -r --contains ${{ github.ref_name }})
             branch=${raw/origin\/}
           fi
-          cd ${{ github.workspace }}; tox -e test-gpu -- $branch
+          cd ${{ github.workspace }}; MERLIN_BRANCH=$branch tox -e test-gpu
+
+  # Single GPU tests
+
+  gpu-tests-conda-cu12:
+    runs-on: linux-amd64-gpu-p100-latest-1
+    container:
+      image: nvidia/cuda:12.1.1-devel-ubuntu22.04
+      env:
+        NVIDIA_VISIBLE_DEVICES: ${{ env.NVIDIA_VISIBLE_DEVICES }}
+    steps:
+      - uses: actions/checkout@v3
+        with:
+          fetch-depth: 0
+      - name: Install Ubuntu packages
+        run: |
+          apt-get update -y
+          apt-get install -y git lsb-release
+      - uses: conda-incubator/setup-miniconda@v2
+        with:
+          miniforge-variant: Mambaforge
+          use-mamba: true
+          activate-environment: cu12-env
+          environment-file: conda/environments/test-cu12.yaml
+          python-version: "3.10"
+      - name: Get Branch name
+        id: get-branch-name
+        uses: NVIDIA-Merlin/.github/actions/branch-name@9f82e25a18e2b4a3f4350e9f287c2c31e906d89e
+      - name: Run tests
+        run: |
+          merlin_branch="${{ steps.get-branch-name.outputs.branch }}"
+          MERLIN_BRANCH=$merlin_branch tox -e test-gpu
+
+  gpu-tests-cu11:
+    runs-on: linux-amd64-gpu-p100-latest-1
+    container:
+      image: nvidia/cuda:11.8.0-devel-ubuntu22.04
+      env:
+        NVIDIA_VISIBLE_DEVICES: ${{ env.NVIDIA_VISIBLE_DEVICES }}
+    steps:
+      - uses: actions/checkout@v3
+        with:
+          fetch-depth: 0
+      - name: Install Ubuntu packages
+        run: |
+          apt-get update -y
+          # libcudnn8 installed for tensorflow GPU support
+          apt-get install -y git lsb-release 'libcudnn8=*cuda11.8'
+      - name: Set up Python 3.8
+        uses: actions/setup-python@v4
+        with:
+          python-version: 3.8
+      - name: Install and upgrade python packages
+        run: |
+          python -m pip install --upgrade pip tox
+      - name: Get Branch name
+        id: get-branch-name
+        uses: NVIDIA-Merlin/.github/actions/branch-name@9f82e25a18e2b4a3f4350e9f287c2c31e906d89e
+      - name: Run tests
+        run: |
+          merlin_branch="${{ steps.get-branch-name.outputs.branch }}"
+          RAPIDS_VERSION=23.04 MERLIN_BRANCH=$merlin_branch tox -e test-gpu-cu11
+
+  gpu-tests-cu12:
+    runs-on: linux-amd64-gpu-p100-latest-1
+    container:
+      image: nvidia/cuda:12.1.1-devel-ubuntu22.04
+      env:
+        NVIDIA_VISIBLE_DEVICES: ${{ env.NVIDIA_VISIBLE_DEVICES }}
+    steps:
+      - uses: actions/checkout@v3
+        with:
+          fetch-depth: 0
+      - name: Install Ubuntu packages
+        run: |
+          apt-get update -y
+          apt-get install -y git lsb-release
+      - name: Set up Python 3.9
+        uses: actions/setup-python@v4
+        with:
+          python-version: 3.9
+      - name: Install and upgrade python packages
+        run: |
+          python -m pip install --upgrade pip tox
+      - name: Get Branch name
+        id: get-branch-name
+        uses: NVIDIA-Merlin/.github/actions/branch-name@9f82e25a18e2b4a3f4350e9f287c2c31e906d89e
+      - name: Run tests
+        run: |
+          merlin_branch="${{ steps.get-branch-name.outputs.branch }}"
+          RAPIDS_VERSION=23.06 MERLIN_BRANCH=$merlin_branch tox -e test-gpu-cu12

conda/environments/test-cu12.yaml

@@ -0,0 +1,11 @@
+name: cu12-env
+channels:
+  - conda-forge
+  - rapidsai-nightly
+  - nvidia
+dependencies:
+  - cuda-version=12
+  - cuda-nvcc=12
+  - cudf=23.08
+  - dask-cudf=23.08
+  - tox=4

requirements/test.txt

@@ -19,7 +19,7 @@ s3fs>=2021.4
 aiobotocore>=1.3.3
 
 # required for synthetic data `merlin.datasets` and notebook tests using merlin models
-merlin-models[tensorflow]@git+https://github.com/NVIDIA-Merlin/models.git
+merlin-models[tensorflow]
 
 # needed to run notebook tests
 nest-asyncio

tests/unit/examples/test_01-Getting-started.py

@@ -23,6 +23,7 @@
 nest_asyncio.apply()
 
 
+@pytest.mark.tensorflow
 def test_example_01_getting_started():
     with testbook(
         REPO_ROOT / "examples" / "01-Getting-started.ipynb",

tests/unit/examples/test_02-Advanced-NVTabular-workflow.py

@@ -27,6 +27,7 @@
 nest_asyncio.apply()
 
 
+@pytest.mark.tensorflow
 def test_example_02_advanced_workflow():
     with testbook(
         REPO_ROOT / "examples" / "02-Advanced-NVTabular-workflow.ipynb",

tox.ini

@@ -39,14 +39,51 @@ sitepackages=true
 ; to install requirements.txt yet. As we get better at python environment isolation, we will
 ; need to add some back.
 deps =
-    pytest
-    pytest-cov
+    -rrequirements/test.txt
+    git+https://github.com/NVIDIA-Merlin/core.git@{env:MERLIN_BRANCH}
+    git+https://github.com/NVIDIA-Merlin/dataloader.git@{env:MERLIN_BRANCH}
+    git+https://github.com/NVIDIA-Merlin/models.git@{env:MERLIN_BRANCH}
 commands =
-    python -m pip install --upgrade git+https://github.com/NVIDIA-Merlin/core.git
-    python -m pip install --upgrade git+https://github.com/NVIDIA-Merlin/dataloader.git
-    python -m pip install --upgrade git+https://github.com/NVIDIA-Merlin/models.git
-    python -m pip install --upgrade git+https://github.com/NVIDIA-Merlin/core.git@{posargs:main}
-    python -m pytest --cov-report term --cov merlin -rxs tests/unit
+    python -m pytest --cov-report term --cov merlin -rxs {posargs:tests/unit}
+
+[testenv:test-gpu-cu11]
+; Runs in: GitHub Actions
+; Runs GPU-based tests.
+setenv =
+    TF_GPU_ALLOCATOR=cuda_malloc_async
+    PIP_EXTRA_INDEX_URL=https://pypi.nvidia.com
+    PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python
+passenv =
+    CUDA_VISIBLE_DEVICES
+deps =
+    -rrequirements/test.txt
+    git+https://github.com/NVIDIA-Merlin/core.git@{env:MERLIN_BRANCH}
+    git+https://github.com/NVIDIA-Merlin/dataloader.git@{env:MERLIN_BRANCH}
+    git+https://github.com/NVIDIA-Merlin/models.git@{env:MERLIN_BRANCH}
+    nvidia-cudnn-cu11==8.6.0.163
+    cudf-cu11=={env:RAPIDS_VERSION}
+    dask-cudf-cu11=={env:RAPIDS_VERSION}
+commands =
+    python -m pytest --cov-report term --cov merlin -rxs -s {posargs:tests/unit}
+
+[testenv:test-gpu-cu12]
+; Runs in: GitHub Actions
+; Runs GPU-based tests.
+setenv =
+    PIP_EXTRA_INDEX_URL=https://pypi.nvidia.com
+    PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python
+passenv =
+    CUDA_VISIBLE_DEVICES
+deps =
+    -rrequirements/test.txt
+    git+https://github.com/NVIDIA-Merlin/core.git@{env:MERLIN_BRANCH}
+    git+https://github.com/NVIDIA-Merlin/dataloader.git@{env:MERLIN_BRANCH}
+    git+https://github.com/NVIDIA-Merlin/models.git@{env:MERLIN_BRANCH}
+    cudf-cu12=={env:RAPIDS_VERSION}
+    dask-cudf-cu12=={env:RAPIDS_VERSION}
+commands =
+    ; Latest TensorFlow PyPI package does not currently support CUDA 12
+    python -m pytest --cov-report term --cov merlin -rxs -s -m 'not tensorflow' {posargs:tests/unit}
 
 [testenv:test-merlin]
 ; Runs in: Internal Jenkins