Migrate vLLM SM, merge rayserve dockerfile, and split PR workflow #57

Workflow file for this run

.github/workflows/pr-vllm-rayserve.yml at 161a30a


	name: PR - vLLM RayServe

	on:
	pull_request:
	branches:
	- main
	paths:
	- "docker/vllm/**"
	- ".github/workflows/pr-vllm-rayserve.yml"

	permissions:
	contents: read

	concurrency:
	group: pr-vllm-rayserve-${{ github.event.pull_request.number }}
	cancel-in-progress: true

	env:
	VLLM_VERSION: 0.10.2

	jobs:
	build-image:
	runs-on:
	- codebuild-runner-${{ github.run_id }}-${{ github.run_attempt }}
	fleet:x86-build-runner
	outputs:
	image-uri: ${{ steps.image-uri-build.outputs.IMAGE_URI }}
	steps:
	- uses: actions/checkout@v5
	- run: .github/scripts/runner_setup.sh
	- run: .github/scripts/buildkitd.sh
	- name: ECR login
	run: \|
	aws ecr get-login-password --region ${{ vars.AWS_REGION }} \| docker login --username AWS --password-stdin ${{ vars.AWS_ACCOUNT_ID }}.dkr.ecr.${{ vars.AWS_REGION }}.amazonaws.com

	- name: Resolve image URI for build
	id: image-uri-build
	run: \|
	IMAGE_URI=${{ vars.AWS_ACCOUNT_ID }}.dkr.ecr.${{ vars.AWS_REGION }}.amazonaws.com/ci:vllm-${{ env.VLLM_VERSION }}-gpu-py312-cu128-ubuntu22.04-rayserve-ec2-pr-${{ github.event.pull_request.number }}
	echo "Image URI to build: ${IMAGE_URI}"
	echo "IMAGE_URI=${IMAGE_URI}" >> ${GITHUB_ENV}
	echo "IMAGE_URI=${IMAGE_URI}" >> ${GITHUB_OUTPUT}

	- name: Build image
	run: \|
	docker buildx build --progress plain \
	--build-arg VLLM_VERSION=${{ env.VLLM_VERSION }} \
	--build-arg CACHE_REFRESH="$(date +"%Y-%m-%d")" \
	--cache-to=type=inline \
	--cache-from=type=registry,ref=${IMAGE_URI} \
	--tag ${IMAGE_URI} \
	--target vllm-rayserve-ec2 \
	-f docker/vllm/Dockerfile .

	- name: Container push
	run: \|
	docker push ${IMAGE_URI}
	docker rmi ${IMAGE_URI}

	regression-test:
	needs: [build-image]
	if: needs.build-image.result == 'success'
	runs-on:
	- codebuild-runner-${{ github.run_id }}-${{ github.run_attempt }}
	fleet:x86-g6xl-runner
	steps:
	- name: Checkout DLC source
	uses: actions/checkout@v5

	- name: Container pull
	uses: ./.github/actions/ecr-authenticate
	with:
	aws_region: ${{ vars.AWS_REGION }}
	aws_account_id: ${{ vars.AWS_ACCOUNT_ID }}
	image_uri: ${{ needs.build-image.outputs.image-uri }}

	- name: Checkout vLLM tests
	uses: actions/checkout@v5
	with:
	repository: vllm-project/vllm
	ref: v${{ env.VLLM_VERSION }}
	path: vllm_source

	- name: Start container
	run: \|
	CONTAINER_ID=$(docker run -d -it --rm --gpus=all --entrypoint /bin/bash \
	-v ${HOME}/.cache/huggingface:/root/.cache/huggingface \
	-v ${HOME}/.cache/vllm:/root/.cache/vllm \
	-v ./vllm_source:/workdir --workdir /workdir \
	-e HF_TOKEN=${{ secrets.HUGGING_FACE_HUB_TOKEN }} \
	${{ needs.build-image.outputs.image-uri }})
	echo "CONTAINER_ID=$CONTAINER_ID" >> $GITHUB_ENV

	- name: Setup for vLLM tests
	run: \|
	docker exec ${CONTAINER_ID} sh -c '
	set -eux
	uv pip install --system -r requirements/common.txt -r requirements/dev.txt --torch-backend=auto
	uv pip install --system pytest pytest-asyncio
	uv pip install --system -e tests/vllm_test_utils
	uv pip install --system hf_transfer
	mkdir src
	mv vllm src/vllm
	'

	- name: Run vLLM tests
	run: \|
	docker exec ${CONTAINER_ID} sh -c '
	set -eux
	nvidia-smi

	# Regression Test # 7min
	cd /workdir/tests
	uv pip install --system modelscope
	pytest -v -s test_regression.py
	'

	- name: Cleanup container and images
	if: always()
	uses: ./.github/actions/container-cleanup
	with:
	container_id: ${{ env.CONTAINER_ID }}

	cuda-test:
	needs: [build-image]
	if: needs.build-image.result == 'success'
	runs-on:
	- codebuild-runner-${{ github.run_id }}-${{ github.run_attempt }}
	fleet:x86-g6xl-runner
	steps:
	- name: Checkout DLC source
	uses: actions/checkout@v5

	- name: Container pull
	uses: ./.github/actions/ecr-authenticate
	with:
	aws_region: ${{ vars.AWS_REGION }}
	aws_account_id: ${{ vars.AWS_ACCOUNT_ID }}
	image_uri: ${{ needs.build-image.outputs.image-uri }}

	- name: Checkout vLLM tests
	uses: actions/checkout@v5
	with:
	repository: vllm-project/vllm
	ref: v${{ env.VLLM_VERSION }}
	path: vllm_source

	- name: Start container
	run: \|
	CONTAINER_ID=$(docker run -d -it --rm --gpus=all --entrypoint /bin/bash \
	-v ${HOME}/.cache/huggingface:/root/.cache/huggingface \
	-v ${HOME}/.cache/vllm:/root/.cache/vllm \
	-v ./vllm_source:/workdir --workdir /workdir \
	-e HF_TOKEN=${{ secrets.HUGGING_FACE_HUB_TOKEN }} \
	${{ needs.build-image.outputs.image-uri }})
	echo "CONTAINER_ID=$CONTAINER_ID" >> $GITHUB_ENV

	- name: Setup for vLLM tests
	run: \|
	docker exec ${CONTAINER_ID} sh -c '
	set -eux
	uv pip install --system -r requirements/common.txt -r requirements/dev.txt --torch-backend=auto
	uv pip install --system pytest pytest-asyncio
	uv pip install --system -e tests/vllm_test_utils
	uv pip install --system hf_transfer
	mkdir src
	mv vllm src/vllm
	'

	- name: Run vLLM tests
	run: \|
	docker exec ${CONTAINER_ID} sh -c '
	set -eux
	nvidia-smi

	# Platform Tests (CUDA) # 4min
	cd /workdir/tests
	pytest -v -s cuda/test_cuda_context.py
	'

	- name: Cleanup container and images
	if: always()
	uses: ./.github/actions/container-cleanup
	with:
	container_id: ${{ env.CONTAINER_ID }}

	example-test:
	needs: [build-image]
	if: needs.build-image.result == 'success'
	runs-on:
	- codebuild-runner-${{ github.run_id }}-${{ github.run_attempt }}
	fleet:x86-g6xl-runner
	steps:
	- name: Checkout DLC source
	uses: actions/checkout@v5

	- name: Container pull
	uses: ./.github/actions/ecr-authenticate
	with:
	aws_region: ${{ vars.AWS_REGION }}
	aws_account_id: ${{ vars.AWS_ACCOUNT_ID }}
	image_uri: ${{ needs.build-image.outputs.image-uri }}

	- name: Checkout vLLM tests
	uses: actions/checkout@v5
	with:
	repository: vllm-project/vllm
	ref: v${{ env.VLLM_VERSION }}
	path: vllm_source

	- name: Start container
	run: \|
	CONTAINER_ID=$(docker run -d -it --rm --gpus=all --entrypoint /bin/bash \
	-v ${HOME}/.cache/huggingface:/root/.cache/huggingface \
	-v ${HOME}/.cache/vllm:/root/.cache/vllm \
	-v ./vllm_source:/workdir --workdir /workdir \
	-e HF_TOKEN=${{ secrets.HUGGING_FACE_HUB_TOKEN }} \
	${{ needs.build-image.outputs.image-uri }})
	echo "CONTAINER_ID=$CONTAINER_ID" >> $GITHUB_ENV

	- name: Setup for vLLM tests
	run: \|
	docker exec ${CONTAINER_ID} sh -c '
	set -eux
	uv pip install --system -r requirements/common.txt -r requirements/dev.txt --torch-backend=auto
	uv pip install --system pytest pytest-asyncio
	uv pip install --system -e tests/vllm_test_utils
	uv pip install --system hf_transfer
	mkdir src
	mv vllm src/vllm
	'

	- name: Run vLLM tests
	run: \|
	docker exec ${CONTAINER_ID} sh -c '
	set -eux
	nvidia-smi

	# Examples Test # 30min
	cd /workdir/examples
	pip install tensorizer # for tensorizer test
	python3 offline_inference/basic/generate.py --model facebook/opt-125m
	# python3 offline_inference/basic/generate.py --model meta-llama/Llama-2-13b-chat-hf --cpu-offload-gb 10
	python3 offline_inference/basic/chat.py
	python3 offline_inference/prefix_caching.py
	python3 offline_inference/llm_engine_example.py
	python3 offline_inference/audio_language.py --seed 0
	python3 offline_inference/vision_language.py --seed 0
	python3 offline_inference/vision_language_pooling.py --seed 0
	python3 offline_inference/vision_language_multi_image.py --seed 0
	VLLM_USE_V1=0 python3 others/tensorize_vllm_model.py --model facebook/opt-125m serialize --serialized-directory /tmp/ --suffix v1 && python3 others/tensorize_vllm_model.py --model facebook/opt-125m deserialize --path-to-tensors /tmp/vllm/facebook/opt-125m/v1/model.tensors
	python3 offline_inference/encoder_decoder_multimodal.py --model-type whisper --seed 0
	python3 offline_inference/basic/classify.py
	python3 offline_inference/basic/embed.py
	python3 offline_inference/basic/score.py
	VLLM_USE_V1=0 python3 offline_inference/profiling.py --model facebook/opt-125m run_num_steps --num-steps 2
	'

	- name: Cleanup container and images
	if: always()
	uses: ./.github/actions/container-cleanup
	with:
	container_id: ${{ env.CONTAINER_ID }}

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Migrate vLLM SM, merge rayserve dockerfile, and split PR workflow #57

Workflow file

Migrate vLLM SM, merge rayserve dockerfile, and split PR workflow #57

Uh oh!

Workflow file for this run