chatai/sglang/.github/workflows/pr-test-amd.yml

120 lines
4.6 KiB
YAML

name: PR Test (AMD)
on:
push:
branches: [ main ]
paths:
- "python/sglang/**"
- "test/**"
- "sgl-kernel/**"
pull_request:
branches: [ main ]
paths:
- "python/sglang/**"
- "test/**"
- "sgl-kernel/**"
workflow_dispatch:
concurrency:
group: pr-test-amd-${{ github.ref }}
cancel-in-progress: true
jobs:
accuracy-test-1-gpu-amd:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
github.event.pull_request.draft == false
runs-on: linux-mi300-gpu-1
steps:
- name: Checkout code
uses: actions/checkout@v4
- name: Setup docker
run: |
# Ensure GPU isolation if pod is part of kubernetes setup with DEVICE_FLAG.
if [ -f "/etc/podinfo/gha-render-devices" ]; then
DEVICE_FLAG=$(cat /etc/podinfo/gha-render-devices)
else
DEVICE_FLAG="--device /dev/dri"
fi
docker pull lmsysorg/sglang:v0.4.3.post4-rocm630
docker run -dt --user root --device=/dev/kfd $DEVICE_FLAG \
-v ${{ github.workspace }}:/sglang-checkout --ipc=host --group-add video \
--cap-add=SYS_PTRACE -e HF_TOKEN=${HF_TOKEN} --security-opt seccomp=unconfined \
-w /sglang-checkout --name ci_sglang \
lmsysorg/sglang:v0.4.3.post4-rocm630
- name: Install dependencies
run: |
docker exec ci_sglang pip install --upgrade pip
docker exec ci_sglang pip uninstall sgl-kernel -y || true
docker exec -w /sglang-checkout/sgl-kernel ci_sglang bash -c "rm -f pyproject.toml && mv pyproject_rocm.toml pyproject.toml && python3 setup_rocm.py install"
docker exec ci_sglang pip install -e "python[dev_hip]"
docker exec -w / ci_sglang git clone https://github.com/merrymercy/human-eval.git
docker exec -w /human-eval ci_sglang pip install -e .
- name: Evaluate Accuracy
timeout-minutes: 20
run: |
docker exec -w /sglang-checkout/test/srt -e SGLANG_IS_IN_CI=1 ci_sglang python3 test_eval_accuracy_large.py
docker exec -w /sglang-checkout/test/srt -e SGLANG_IS_IN_CI=1 ci_sglang python3 test_eval_fp8_accuracy.py
docker exec -w /sglang-checkout/test/srt -e SGLANG_IS_IN_CI=1 ci_sglang python3 models/test_qwen_models.py
mla-test-1-gpu-amd:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
github.event.pull_request.draft == false
runs-on: linux-mi300-gpu-1
steps:
- name: Checkout code
uses: actions/checkout@v4
- name: Setup docker
run: |
# Ensure GPU isolation if pod is part of kubernetes setup with DEVICE_FLAG.
if [ -f "/etc/podinfo/gha-render-devices" ]; then
DEVICE_FLAG=$(cat /etc/podinfo/gha-render-devices)
else
DEVICE_FLAG="--device /dev/dri"
fi
docker pull lmsysorg/sglang:v0.4.3.post4-rocm630
docker run -dt --user root --device=/dev/kfd $DEVICE_FLAG \
-v ${{ github.workspace }}:/sglang-checkout --ipc=host --group-add video \
--cap-add=SYS_PTRACE -e HF_TOKEN=${{ secrets.AMD_HF_TOKEN }} --security-opt seccomp=unconfined \
-w /sglang-checkout --name ci_sglang \
lmsysorg/sglang:v0.4.3.post4-rocm630
- name: Install dependencies
run: |
docker exec ci_sglang pip install --upgrade pip
docker exec ci_sglang pip uninstall sgl-kernel -y || true
docker exec -w /sglang-checkout/sgl-kernel ci_sglang bash -c "rm -f pyproject.toml && mv pyproject_rocm.toml pyproject.toml && python3 setup_rocm.py install"
docker exec ci_sglang pip install -e "python[dev_hip]"
docker exec ci_sglang pip install py-spy || true
docker exec -w / ci_sglang git clone https://github.com/merrymercy/human-eval.git
docker exec -w /human-eval ci_sglang pip install -e .
- name: MLA TEST
timeout-minutes: 20
run: |
docker exec -w /sglang-checkout/test/srt -e SGLANG_IS_IN_CI=1 ci_sglang python3 test_mla.py
finish:
if: always()
needs: [
accuracy-test-1-gpu-amd, mla-test-1-gpu-amd
]
runs-on: ubuntu-latest
steps:
- name: Check all dependent job statuses
run: |
results=(${{ join(needs.*.result, ' ') }})
for result in "${results[@]}"; do
if [ "$result" = "failure" ] || [ "$result" = "cancelled" ]; then
echo "Job failed with result: $result"
exit 1
fi
done
echo "All jobs completed successfully"
exit 0