81 lines
2.2 KiB
YAML
81 lines
2.2 KiB
YAML
name: Rerun UT
|
|
run-name: ${{ inputs.pr_head_sha && format('[rerun-ut] {0}', inputs.pr_head_sha) || '[rerun-ut]' }}
|
|
|
|
on:
|
|
workflow_dispatch:
|
|
inputs:
|
|
test_command:
|
|
description: "Test command to run (e.g. 'registered/core/test_srt_endpoint.py TestSRTEndpoint.test_simple_decode')"
|
|
required: true
|
|
type: string
|
|
runner_label:
|
|
description: "Runner label"
|
|
required: true
|
|
type: choice
|
|
options:
|
|
- 1-gpu-h100
|
|
- 1-gpu-5090
|
|
- 2-gpu-h100
|
|
- 4-gpu-h100
|
|
- 4-gpu-a10
|
|
- 4-gpu-b200
|
|
- 8-gpu-h200
|
|
- 8-gpu-h20
|
|
- 8-gpu-b200
|
|
pr_head_sha:
|
|
description: "PR head SHA to checkout (for /rerun-ut on fork PRs)"
|
|
required: false
|
|
type: string
|
|
default: ""
|
|
use_deepep:
|
|
description: "Use ci_install_deepep.sh instead of ci_install_dependency.sh"
|
|
required: false
|
|
type: string
|
|
default: "false"
|
|
|
|
env:
|
|
SGLANG_IS_IN_CI: true
|
|
SGLANG_CUDA_COREDUMP: "1"
|
|
SGLANG_JIT_DEEPGEMM_FAST_WARMUP: true
|
|
|
|
permissions:
|
|
actions: write
|
|
contents: read
|
|
|
|
jobs:
|
|
rerun-ut-cuda:
|
|
runs-on: ${{ inputs.runner_label }}
|
|
timeout-minutes: 120
|
|
env:
|
|
RUNNER_LABELS: ${{ inputs.runner_label }}
|
|
SGLANG_CI_RDMA_ALL_DEVICES: ${{ inputs.runner_label == '8-gpu-h20' && 'mlx5_1,mlx5_2,mlx5_3,mlx5_4' || '' }}
|
|
steps:
|
|
- name: Checkout code
|
|
uses: actions/checkout@v4
|
|
with:
|
|
ref: ${{ inputs.pr_head_sha || github.sha }}
|
|
|
|
- name: Install dependencies
|
|
timeout-minutes: 20
|
|
run: |
|
|
if [[ "${{ inputs.runner_label }}" == "1-gpu-5090" ]]; then
|
|
source /etc/profile.d/sglang-ci.sh
|
|
fi
|
|
if [[ "${{ inputs.use_deepep }}" == "true" ]]; then
|
|
bash scripts/ci/cuda/ci_install_deepep.sh
|
|
else
|
|
bash scripts/ci/cuda/ci_install_dependency.sh
|
|
fi
|
|
|
|
- name: Run test
|
|
timeout-minutes: 60
|
|
run: |
|
|
if [[ "${{ inputs.runner_label }}" == "1-gpu-5090" ]]; then
|
|
source /etc/profile.d/sglang-ci.sh
|
|
fi
|
|
cd test/
|
|
python3 ${{ inputs.test_command }}
|
|
|
|
- uses: ./.github/actions/upload-cuda-coredumps
|
|
if: always()
|