Skip to content

TorchBench Nightly Docker Build #491

TorchBench Nightly Docker Build

TorchBench Nightly Docker Build #491

name: TorchBench nightly docker build
on:
schedule:
# Push the nightly docker daily at 3 PM UTC
- cron: '0 15 * * *'
workflow_dispatch:
inputs:
nightly_date:
description: "PyTorch nightly version"
required: false
env:
WITH_PUSH: "true"
CONDA_ENV: "torchbench"
DOCKER_IMAGE: "ghcr.io/pytorch/torchbench:latest"
SETUP_SCRIPT: "/workspace/setup_instance.sh"
HUGGING_FACE_HUB_TOKEN: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
jobs:
build-push-docker:
if: ${{ github.repository_owner == 'pytorch' }}
runs-on: [self-hosted, linux.4xlarge]
environment: docker-s3-upload
steps:
- name: Checkout
uses: actions/checkout@v3
with:
path: benchmark
- name: Login to GitHub Container Registry
if: ${{ env.WITH_PUSH == 'true' }}
uses: docker/login-action@v2
with:
registry: ghcr.io
username: pytorch
password: ${{ secrets.TORCHBENCH_ACCESS_TOKEN }}
- name: Build TorchBench nightly docker
run: |
export NIGHTLY_DATE="${{ github.event.inputs.nightly_date }}"
if [ -z "${NIGHTLY_DATE}" ]; then
export TODAY=$(date +'%Y%m%d')
export DOCKER_TAG=dev${TODAY}
else
export DOCKER_TAG=dev${NIGHTLY_DATE}
fi
cd benchmark/docker
full_ref="${{ github.ref }}"
prefix="refs/heads/"
branch_name=${full_ref#$prefix}
docker build . --build-arg TORCHBENCH_BRANCH="${branch_name}" --build-arg FORCE_DATE="${NIGHTLY_DATE}" \
-f torchbench-nightly.dockerfile -t ghcr.io/pytorch/torchbench:${DOCKER_TAG}
docker tag ghcr.io/pytorch/torchbench:${DOCKER_TAG} ghcr.io/pytorch/torchbench:latest
- name: Push docker to remote
if: ${{ env.WITH_PUSH == 'true' }}
run: |
export TODAY=$(date +'%Y%m%d')
export DOCKER_TAG=dev${TODAY}
docker push ghcr.io/pytorch/torchbench:${DOCKER_TAG}
docker push ghcr.io/pytorch/torchbench:latest
test-docker:
if: ${{ github.repository_owner == 'pytorch' }}
runs-on: [self-hosted, linux.g5.4xlarge.nvidia.gpu]
needs: build-push-docker
environment: docker-s3-upload
steps:
- name: "[FB EMPLOYEES] Enable SSH (Click me for login details)"
uses: pytorch/test-infra/.github/actions/setup-ssh@main
with:
github-secret: ${{ secrets.TORCHBENCH_ACCESS_TOKEN }}
- name: Checkout
uses: actions/checkout@v3
with:
path: benchmark
- name: Pull docker image
uses: pytorch/test-infra/.github/actions/pull-docker-image@main
with:
docker-image: ${{ env.DOCKER_IMAGE }}
- name: Install NVIDIA Driver, docker runtime, set GPU_FLAG
id: install-nvidia-driver
uses: pytorch/test-infra/.github/actions/setup-nvidia@main
- name: Install and Test TorchBench Docker
run: |
container_name=$(docker run \
-e CONDA_ENV="${CONDA_ENV}" \
-e SETUP_SCRIPT="${SETUP_SCRIPT}" \
-e HUGGING_FACE_HUB_TOKEN="${HUGGING_FACE_HUB_TOKEN}" \
--tty \
--detach \
--shm-size=32gb \
-v "${PWD}/benchmark:/benchmark" \
--gpus all \
-w / \
"${{ env.DOCKER_IMAGE }}" \
tail -f /dev/null
)
echo "Container name: ${container_name}"
docker exec -t -w "/" "${container_name}" bash -c "sudo chown -R runner /benchmark; sudo chgrp -R runner /benchmark"
docker exec -t -w "/benchmark" "${container_name}" bash /benchmark/.ci/tritonbench/test.sh
- name: Teardown Linux
uses: pytorch/test-infra/.github/actions/teardown-linux@main
if: always()
concurrency:
group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}-${{ github.event_name == 'workflow_dispatch' }}
cancel-in-progress: true