feat: Port SGLang from v1 to v2 #44
Workflow file for this run
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| name: PR - SGLang | |
| on: | |
| pull_request: | |
| branches: | |
| - main | |
| paths: | |
| - "docker/sglang/**" | |
| permissions: | |
| contents: read | |
| concurrency: | |
| group: pr-sglang-${{ github.event.pull_request.number }} | |
| cancel-in-progress: true | |
| jobs: | |
| check-changes: | |
| runs-on: ubuntu-latest | |
| outputs: | |
| sglang-sagemaker: ${{ steps.changes.outputs.sglang-sagemaker }} | |
| steps: | |
| - uses: actions/checkout@v5 | |
| - uses: actions/setup-python@v6 | |
| with: | |
| python-version: "3.12" | |
| - uses: pre-commit/action@v3.0.1 | |
| with: | |
| extra_args: --all-files | |
| - name: Detect file changes | |
| id: changes | |
| uses: dorny/paths-filter@v3 | |
| with: | |
| filters: | | |
| sglang-sagemaker: | |
| - "docker/sglang/Dockerfile" | |
| build-sglang-image: | |
| needs: [check-changes] | |
| if: needs.check-changes.outputs.sglang-sagemaker == 'true' | |
| runs-on: | |
| - codebuild-runner-${{ github.run_id }}-${{ github.run_attempt }} | |
| fleet:x86-build-runner | |
| outputs: | |
| my_output: ${{ 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 ${{ secrets.AWS_REGION }} | docker login --username AWS --password-stdin ${{ secrets.AWS_ACCOUNT_ID }}.dkr.ecr.${{ secrets.AWS_REGION }}.amazonaws.com | |
| - name: Resolve image URI for build | |
| id: image_uri_build | |
| run: | | |
| IMAGE_URI=${{ secrets.AWS_ACCOUNT_ID }}.dkr.ecr.${{ secrets.AWS_REGION }}.amazonaws.com/ci:sglang-0.5.5-gpu-py312-cu129-ubuntu22.04-sagemaker-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" >> $GITHUB_OUTPUT | |
| - name: Build image | |
| run: | | |
| docker buildx build --progress plain \ | |
| --build-arg CACHE_REFRESH="$(date +"%Y-%m-%d")" \ | |
| --cache-to=type=inline \ | |
| --cache-from=type=registry,ref=$IMAGE_URI \ | |
| --tag $IMAGE_URI \ | |
| --target sglang-sagemaker \ | |
| -f docker/sglang/Dockerfile . | |
| - name: Docker push and save image URI artifact | |
| run: | | |
| docker push $IMAGE_URI | |
| docker rmi $IMAGE_URI | |
| echo $IMAGE_URI > image_uri.txt | |
| - name: Upload image URI artifact | |
| uses: actions/upload-artifact@v4 | |
| with: | |
| name: sglang-sagemaker-image-uri | |
| path: image_uri.txt | |
| sglang-local-benchmark-test: | |
| needs: [build-sglang-image] | |
| if: needs.build-sglang-image.result == 'success' | |
| runs-on: | |
| - codebuild-runner-${{ github.run_id }}-${{ github.run_attempt }} | |
| fleet:x86-g6xl-runner | |
| steps: | |
| - name: Checkout DLC source | |
| uses: actions/checkout@v5 | |
| - name: echo container | |
| run: | | |
| echo ${{ needs.build-sglang-image.outputs.my_output }} | |
| - name: Container pull | |
| uses: ./.github/actions/container-pull | |
| with: | |
| aws_region: ${{ secrets.AWS_REGION }} | |
| aws_account_id: ${{ secrets.AWS_ACCOUNT_ID }} | |
| image_uri: ${{ needs.build-sglang-image.outputs.my_output }} | |
| - name: Setup for SGLang datasets | |
| run: | | |
| mkdir -p /tmp/sglang/dataset | |
| if [ ! -f /tmp/sglang/dataset/ShareGPT_V3_unfiltered_cleaned_split.json ]; then | |
| echo "Downloading ShareGPT dataset..." | |
| wget -P /tmp/sglang/dataset https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered/resolve/main/ShareGPT_V3_unfiltered_cleaned_split.json | |
| else | |
| echo "ShareGPT dataset already exists. Skipping download." | |
| fi | |
| - name: Start container | |
| run: | | |
| CONTAINER_ID=$(docker run -d -it --rm --gpus=all \ | |
| -v ${HOME}/.cache/huggingface:/root/.cache/huggingface \ | |
| -v /tmp/sglang/dataset:/dataset \ | |
| -v ./sglang_source:/workdir --workdir /workdir \ | |
| -p 30000:30000 \ | |
| -e SM_SGLANG_MODEL_PATH=Qwen/Qwen3-0.6B \ | |
| -e SM_SGLANG_REASONING_PARSER=qwen3 \ | |
| -e SM_SGLANG_HOST=127.0.0.1 \ | |
| -e SM_SGLANG_PORT=30000 \ | |
| -e HF_TOKEN=${{ secrets.HUGGING_FACE_HUB_TOKEN }} \ | |
| ${IMAGE_URI}) | |
| echo "CONTAINER_ID=$CONTAINER_ID" >> $GITHUB_ENV | |
| echo "Waiting for container startup ..." | |
| sleep 300s | |
| docker logs ${CONTAINER_ID} | |
| - name: Run SGLang tests | |
| run: | | |
| docker exec ${CONTAINER_ID} python3 -m sglang.bench_serving \ | |
| --backend sglang \ | |
| --host 127.0.0.1 --port 30000 \ | |
| --num-prompts 1000 \ | |
| --model Qwen/Qwen3-0.6B \ | |
| --dataset-name sharegpt \ | |
| --dataset-path /dataset/ShareGPT_V3_unfiltered_cleaned_split.json | |
| - name: Cleanup SGLang datasets | |
| run: | | |
| rm -rf /tmp/sglang/dataset | |
| - name: Cleanup container and images | |
| if: always() | |
| uses: ./.github/actions/container-cleanup | |
| with: | |
| container_id: ${{ env.CONTAINER_ID }} |