chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

[AMD] CI - Add MI35x nightly/PR tests for kv-cache-fp8 and allreduce-fusion (DeepSeek) (#19834)

Browse Source 
Co-authored-by: bingxche <Bingxu.Chen@amd.com>
This commit is contained in:
YC Tseng
2026-03-05 18:09:57 +03:00
committed by GitHub
parent 0de0d74195
commit b5edab57f2
13 changed files with 1614 additions and 177 deletions
Download Patch File Download Diff File Expand all files Collapse all files

221
.github/workflows/nightly-test-amd-rocm720.yml vendored
View File

@@ -21,46 +21,10 @@ on:
type: boolean
default: true
job_filter:
description: 'Select which job to run (leave empty or "all" to run all jobs)'
description: 'Comma-separated list of jobs to run (e.g. "nightly-8-gpu-grok2-rocm720,nightly-8-gpu-deepseek-v31-rocm720"). Leave empty or "all" to run all jobs.'
required: false
type: choice
type: string
default: 'all'
options:
- 'all'
# MI30x ROCm 7.2 Unit Tests
- 'nightly-test-1-gpu-unit-rocm720'
# MI30x ROCm 7.2 Accuracy Tests (GSM8K / MMMU)
- 'nightly-accuracy-2-gpu-rocm720'
- 'nightly-accuracy-2-gpu-vlm-rocm720'
- 'nightly-perf-2-gpu-text-rocm720'
- 'nightly-perf-2-gpu-vlm-rocm720'
- 'nightly-accuracy-8-gpu-rocm720'
# MI30x ROCm 7.2 Accuracy + Performance Tests (combined)
- 'nightly-8-gpu-grok1-int4-rocm720'
- 'nightly-8-gpu-grok2-rocm720'
- 'nightly-8-gpu-deepseek-v31-rocm720'
- 'nightly-8-gpu-deepseek-v32-rocm720'
- 'nightly-8-gpu-deepseek-v32-mtp-rocm720'
- 'nightly-8-gpu-kimi-k25-rocm720'
- 'nightly-8-gpu-qwen3-235b-rocm720'
- 'nightly-8-gpu-qwen35-rocm720'
- 'nightly-8-gpu-glm5-rocm720'
- 'nightly-8-gpu-minimax-m25-rocm720'
# MI35x ROCm 7.2 jobs
- 'nightly-test-1-gpu-mi35x-rocm720'
- 'nightly-8-gpu-mi35x-qwen3-235b-mxfp4-rocm720'
- 'nightly-8-gpu-mi35x-qwen35-rocm720'
- 'nightly-accuracy-8-gpu-mi35x-rocm720'
- 'nightly-8-gpu-mi35x-grok1-int4-rocm720'
- 'nightly-8-gpu-mi35x-grok2-rocm720'
- 'nightly-8-gpu-mi35x-deepseek-r1-mxfp4-rocm720'
- 'nightly-accuracy-8-gpu-mi35x-deepseek-v32-rocm720'
- 'nightly-accuracy-8-gpu-mi35x-deepseek-v32-mtp-rocm720'
- 'nightly-perf-8-gpu-mi35x-deepseek-v32-basic-rocm720'
- 'nightly-perf-8-gpu-mi35x-deepseek-v32-mtp-rocm720'
- 'nightly-8-gpu-mi35x-kimi-k25-rocm720'
- 'nightly-8-gpu-mi35x-glm5-rocm720'
- 'nightly-8-gpu-mi35x-minimax-m25-rocm720'
workflow_call:
inputs:
ref:
@@ -98,7 +62,7 @@ jobs:
# ============================================== MI30x ROCm 7.2 Unit Tests ==============================================
# 1-GPU Unit Tests - LoRA, debug utils, scheduler, etc. (MI30x ROCm 7.2)
nightly-test-1-gpu-unit-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-test-1-gpu-unit-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-test-1-gpu-unit-rocm720,'))
runs-on: linux-mi325-1gpu-sglang
steps:
- name: Checkout code
@@ -127,7 +91,7 @@ jobs:
# ============================================== MI30x ROCm 7.2 Accuracy Tests ==============================================
# 2-GPU Accuracy Tests - GSM8K eval (MI30x ROCm 7.2)
nightly-accuracy-2-gpu-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-accuracy-2-gpu-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-accuracy-2-gpu-rocm720,'))
runs-on: linux-mi325-2gpu-sglang
steps:
- name: Checkout code
@@ -155,7 +119,7 @@ jobs:
# 2-GPU VLM Accuracy Tests - Vision-Language Models MMMU evaluation (ROCm 7.2)
nightly-accuracy-2-gpu-vlm-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-accuracy-2-gpu-vlm-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-accuracy-2-gpu-vlm-rocm720,'))
runs-on: linux-mi325-2gpu-sglang
steps:
- name: Checkout code
@@ -184,7 +148,7 @@ jobs:
# 2-GPU Text Models Performance Tests (ROCm 7.2)
nightly-perf-2-gpu-text-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-perf-2-gpu-text-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-perf-2-gpu-text-rocm720,'))
runs-on: linux-mi325-2gpu-sglang
steps:
- name: Checkout code
@@ -214,7 +178,7 @@ jobs:
# 2-GPU VLM Performance Tests (ROCm 7.2)
nightly-perf-2-gpu-vlm-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-perf-2-gpu-vlm-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-perf-2-gpu-vlm-rocm720,'))
runs-on: linux-mi325-2gpu-sglang
steps:
- name: Checkout code
@@ -244,7 +208,7 @@ jobs:
# 8-GPU Accuracy Tests - GPT-OSS, Grok1-FP8 (ROCm 7.2)
nightly-accuracy-8-gpu-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-accuracy-8-gpu-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-accuracy-8-gpu-rocm720,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
@@ -284,7 +248,7 @@ jobs:
# ============================================== MI30x ROCm 7.2 Combined Accuracy + Performance Tests ==============================================
# 8-GPU Grok1-INT4 (Accuracy + Performance) ROCm 7.2
nightly-8-gpu-grok1-int4-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-grok1-int4-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-grok1-int4-rocm720,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
@@ -327,7 +291,7 @@ jobs:
# 8-GPU Grok2 (Accuracy + Performance) ROCm 7.2
nightly-8-gpu-grok2-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-grok2-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-grok2-rocm720,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
@@ -370,7 +334,7 @@ jobs:
# 8-GPU DeepSeek-V3.1 (Accuracy + Performance) ROCm 7.2
nightly-8-gpu-deepseek-v31-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-deepseek-v31-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-deepseek-v31-rocm720,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
@@ -413,7 +377,7 @@ jobs:
# 8-GPU DeepSeek-V3.2 (Basic Accuracy + Perf) ROCm 7.2
nightly-8-gpu-deepseek-v32-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-deepseek-v32-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-deepseek-v32-rocm720,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
@@ -454,7 +418,7 @@ jobs:
# 8-GPU DeepSeek-V3.2 MTP (MTP Accuracy + Perf) ROCm 7.2
nightly-8-gpu-deepseek-v32-mtp-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-deepseek-v32-mtp-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-deepseek-v32-mtp-rocm720,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
@@ -493,9 +457,39 @@ jobs:
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
exit ${TEST_EXIT_CODE:-0}
# 8-GPU DeepSeek-V3 KV FP8 (Basic + MTP with --kv-cache-dtype fp8_e4m3) ROCm 7.2
nightly-8-gpu-deepseek-v3-kv-fp8-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-deepseek-v3-kv-fp8-rocm720,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
uses: actions/checkout@v4
with:
ref: ${{ inputs.ref || github.ref }}
- name: Setup docker (ROCm 7.2)
run: |
touch github_summary.md
bash scripts/ci/amd/amd_ci_start_container.sh --rocm-version rocm720
env:
GITHUB_WORKSPACE: ${{ github.workspace }}
- name: Install dependencies
run: bash scripts/ci/amd/amd_ci_install_dependency.sh --skip-test-time-deps
- name: DeepSeek-V3 KV FP8 Test ROCm 7.2 (8-GPU Basic + MTP)
timeout-minutes: 120
run: |
> github_summary.md # Clear summary file
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
python3 run_suite.py --hw amd --suite nightly-amd-8-gpu-deepseek-v3-kv-fp8 --nightly --timeout-per-file 3600 ${{ inputs.continue_on_error && '--continue-on-error' || '' }} || TEST_EXIT_CODE=$?
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
exit ${TEST_EXIT_CODE:-0}
# 8-GPU Kimi-K2.5 (Accuracy) ROCm 7.2
nightly-8-gpu-kimi-k25-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-kimi-k25-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-kimi-k25-rocm720,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
@@ -525,7 +519,7 @@ jobs:
# 8-GPU Qwen3-235B (Accuracy + Performance) ROCm 7.2
nightly-8-gpu-qwen3-235b-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-qwen3-235b-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-qwen3-235b-rocm720,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
@@ -555,7 +549,7 @@ jobs:
# 8-GPU Qwen 3.5 (Accuracy) ROCm 7.2
nightly-8-gpu-qwen35-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-qwen35-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-qwen35-rocm720,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
@@ -587,7 +581,7 @@ jobs:
# 8-GPU GLM-5 (Accuracy) ROCm 7.2
nightly-8-gpu-glm5-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-glm5-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-glm5-rocm720,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
@@ -620,7 +614,7 @@ jobs:
# 8-GPU MiniMax-M2.5 (Accuracy) ROCm 7.2
nightly-8-gpu-minimax-m25-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-minimax-m25-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-minimax-m25-rocm720,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
@@ -652,7 +646,7 @@ jobs:
# ============================================== MI35x ROCm 7.2 Tests ==============================================
# MI35x 1-GPU ROCm 7.2 tests
nightly-test-1-gpu-mi35x-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-test-1-gpu-mi35x-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-test-1-gpu-mi35x-rocm720,'))
runs-on: linux-mi35x-gpu-1
steps:
- name: Checkout code
@@ -681,7 +675,7 @@ jobs:
# MI35x 8-GPU Accuracy Tests - GPT-OSS (ROCm 7.2)
nightly-accuracy-8-gpu-mi35x-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-accuracy-8-gpu-mi35x-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-accuracy-8-gpu-mi35x-rocm720,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -713,7 +707,7 @@ jobs:
# MI35x 8-GPU Grok1-INT4 (Accuracy + Performance) ROCm 7.2
nightly-8-gpu-mi35x-grok1-int4-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-mi35x-grok1-int4-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-mi35x-grok1-int4-rocm720,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -759,7 +753,7 @@ jobs:
# MI35x 8-GPU Grok2 (Accuracy + Performance) ROCm 7.2
nightly-8-gpu-mi35x-grok2-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-mi35x-grok2-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-mi35x-grok2-rocm720,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -805,7 +799,7 @@ jobs:
# MI35x 8-GPU DeepSeek-R1-MXFP4 (Accuracy + Performance) ROCm 7.2
nightly-8-gpu-mi35x-deepseek-r1-mxfp4-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-mi35x-deepseek-r1-mxfp4-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-mi35x-deepseek-r1-mxfp4-rocm720,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -847,9 +841,97 @@ jobs:
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
exit ${TEST_EXIT_CODE:-0}
# MI35x 8-GPU DeepSeek-R1-MXFP4 KV FP8 (Accuracy + Performance) ROCm 7.2
nightly-8-gpu-mi35x-deepseek-r1-mxfp4-kv-fp8-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-mi35x-deepseek-r1-mxfp4-kv-fp8-rocm720,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
uses: actions/checkout@v4
with:
ref: ${{ inputs.ref || github.ref }}
- name: Setup docker (ROCm 7.2)
run: |
touch github_summary.md
bash scripts/ci/amd/amd_ci_start_container.sh --rocm-version rocm720
env:
GITHUB_WORKSPACE: ${{ github.workspace }}
- name: Install dependencies
run: |
bash scripts/ci/amd/amd_ci_install_dependency.sh --skip-test-time-deps
# Install tabulate for run_suite.py (missing in MI35x container)
bash scripts/ci/amd/amd_ci_exec.sh pip install tabulate
- name: Accuracy Test MI35x ROCm 7.2 (8-GPU DeepSeek-R1-MXFP4 KV FP8)
timeout-minutes: 180
run: |
> github_summary.md # Clear summary file
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
python3 run_suite.py --hw amd --suite nightly-amd-8-gpu-mi35x-deepseek-r1-mxfp4-kv-fp8 --nightly --timeout-per-file 7200 ${{ inputs.continue_on_error && '--continue-on-error' || '' }} || TEST_EXIT_CODE=$?
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
exit ${TEST_EXIT_CODE:-0}
- name: Performance Test MI35x ROCm 7.2 (8-GPU DeepSeek-R1-MXFP4 KV FP8)
timeout-minutes: 300
continue-on-error: true
run: |
> github_summary.md # Clear summary file
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
python3 registered/amd/perf/mi35x/test_deepseek_r1_mxfp4_kv_fp8_perf_mi35x.py || TEST_EXIT_CODE=$?
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
exit ${TEST_EXIT_CODE:-0}
# MI35x 8-GPU DeepSeek-R1-MXFP4 AllReduce Fusion (Accuracy + Performance) ROCm 7.2
nightly-8-gpu-mi35x-deepseek-r1-mxfp4-ar-fusion-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-mi35x-deepseek-r1-mxfp4-ar-fusion-rocm720,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
uses: actions/checkout@v4
with:
ref: ${{ inputs.ref || github.ref }}
- name: Setup docker (ROCm 7.2)
run: |
touch github_summary.md
bash scripts/ci/amd/amd_ci_start_container.sh --rocm-version rocm720
env:
GITHUB_WORKSPACE: ${{ github.workspace }}
- name: Install dependencies
run: |
bash scripts/ci/amd/amd_ci_install_dependency.sh --skip-test-time-deps
# Install tabulate for run_suite.py (missing in MI35x container)
bash scripts/ci/amd/amd_ci_exec.sh pip install tabulate
- name: Accuracy Test MI35x ROCm 7.2 (8-GPU DeepSeek-R1-MXFP4 AllReduce Fusion)
timeout-minutes: 180
run: |
> github_summary.md # Clear summary file
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
python3 run_suite.py --hw amd --suite nightly-amd-8-gpu-mi35x-deepseek-r1-mxfp4-ar-fusion --nightly --timeout-per-file 7200 ${{ inputs.continue_on_error && '--continue-on-error' || '' }} || TEST_EXIT_CODE=$?
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
exit ${TEST_EXIT_CODE:-0}
- name: Performance Test MI35x ROCm 7.2 (8-GPU DeepSeek-R1-MXFP4 AllReduce Fusion)
timeout-minutes: 300
continue-on-error: true
run: |
> github_summary.md # Clear summary file
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
python3 registered/amd/perf/mi35x/test_deepseek_r1_mxfp4_ar_fusion_perf_mi35x.py || TEST_EXIT_CODE=$?
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
exit ${TEST_EXIT_CODE:-0}
# MI35x 8-GPU DeepSeek-V3.2 Accuracy Test (ROCm 7.2)
nightly-accuracy-8-gpu-mi35x-deepseek-v32-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-accuracy-8-gpu-mi35x-deepseek-v32-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-accuracy-8-gpu-mi35x-deepseek-v32-rocm720,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -882,7 +964,7 @@ jobs:
# MI35x 8-GPU DeepSeek-V3.2 TP+MTP Accuracy Test (ROCm 7.2)
nightly-accuracy-8-gpu-mi35x-deepseek-v32-mtp-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-accuracy-8-gpu-mi35x-deepseek-v32-mtp-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-accuracy-8-gpu-mi35x-deepseek-v32-mtp-rocm720,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -915,7 +997,7 @@ jobs:
# MI35x 8-GPU DeepSeek-V3.2 Performance Test (Basic) ROCm 7.2
nightly-perf-8-gpu-mi35x-deepseek-v32-basic-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-perf-8-gpu-mi35x-deepseek-v32-basic-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-perf-8-gpu-mi35x-deepseek-v32-basic-rocm720,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -948,7 +1030,7 @@ jobs:
# MI35x 8-GPU Kimi-K2.5 (Accuracy) ROCm 7.2
nightly-8-gpu-mi35x-kimi-k25-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-mi35x-kimi-k25-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-mi35x-kimi-k25-rocm720,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -981,7 +1063,7 @@ jobs:
# MI35x 8-GPU Qwen3-235B-MXFP4 (Accuracy + Performance) ROCm 7.2
nightly-8-gpu-mi35x-qwen3-235b-mxfp4-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-mi35x-qwen3-235b-mxfp4-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-mi35x-qwen3-235b-mxfp4-rocm720,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -1014,7 +1096,7 @@ jobs:
# MI35x 8-GPU Qwen 3.5 (Accuracy) ROCm 7.2
nightly-8-gpu-mi35x-qwen35-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-mi35x-qwen35-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-mi35x-qwen35-rocm720,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -1046,7 +1128,7 @@ jobs:
exit ${TEST_EXIT_CODE:-0}
nightly-8-gpu-mi35x-glm5-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-mi35x-glm5-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-mi35x-glm5-rocm720,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -1081,7 +1163,7 @@ jobs:
# MI35x 8-GPU MiniMax-M2.5 (Accuracy) ROCm 7.2
nightly-8-gpu-mi35x-minimax-m25-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-mi35x-minimax-m25-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-mi35x-minimax-m25-rocm720,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -1114,7 +1196,7 @@ jobs:
# MI35x 8-GPU DeepSeek-V3.2 Performance Test (MTP) ROCm 7.2
nightly-perf-8-gpu-mi35x-deepseek-v32-mtp-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-perf-8-gpu-mi35x-deepseek-v32-mtp-rocm720')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-perf-8-gpu-mi35x-deepseek-v32-mtp-rocm720,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -1163,6 +1245,7 @@ jobs:
- nightly-8-gpu-deepseek-v31-rocm720
- nightly-8-gpu-deepseek-v32-rocm720
- nightly-8-gpu-deepseek-v32-mtp-rocm720
- nightly-8-gpu-deepseek-v3-kv-fp8-rocm720
- nightly-8-gpu-kimi-k25-rocm720
- nightly-8-gpu-qwen3-235b-rocm720
- nightly-8-gpu-qwen35-rocm720
@@ -1174,6 +1257,8 @@ jobs:
- nightly-8-gpu-mi35x-grok1-int4-rocm720
- nightly-8-gpu-mi35x-grok2-rocm720
- nightly-8-gpu-mi35x-deepseek-r1-mxfp4-rocm720
- nightly-8-gpu-mi35x-deepseek-r1-mxfp4-kv-fp8-rocm720
- nightly-8-gpu-mi35x-deepseek-r1-mxfp4-ar-fusion-rocm720
- nightly-accuracy-8-gpu-mi35x-deepseek-v32-rocm720
- nightly-accuracy-8-gpu-mi35x-deepseek-v32-mtp-rocm720
- nightly-perf-8-gpu-mi35x-deepseek-v32-basic-rocm720

221
.github/workflows/nightly-test-amd.yml vendored
View File

@@ -21,46 +21,10 @@ on:
type: boolean
default: true
job_filter:
description: 'Select which job to run (leave empty or "all" to run all jobs)'
description: 'Comma-separated list of jobs to run (e.g. "nightly-8-gpu-grok2,nightly-8-gpu-deepseek-v31"). Leave empty or "all" to run all jobs.'
required: false
type: choice
type: string
default: 'all'
options:
- 'all'
# MI30x Unit Tests
- 'nightly-test-1-gpu-unit'
# MI30x Accuracy Tests (GSM8K / MMMU)
- 'nightly-accuracy-2-gpu'
- 'nightly-accuracy-2-gpu-vlm'
- 'nightly-perf-2-gpu-text'
- 'nightly-perf-2-gpu-vlm'
- 'nightly-accuracy-8-gpu'
# MI30x Accuracy + Performance Tests (combined)
- 'nightly-8-gpu-grok1-int4'
- 'nightly-8-gpu-grok2'
- 'nightly-8-gpu-deepseek-v31'
- 'nightly-8-gpu-deepseek-v32'
- 'nightly-8-gpu-deepseek-v32-mtp'
- 'nightly-8-gpu-kimi-k25'
- 'nightly-8-gpu-qwen3-235b'
- 'nightly-8-gpu-qwen35'
- 'nightly-8-gpu-glm5'
- 'nightly-8-gpu-minimax-m25'
# MI35x jobs
- 'nightly-test-1-gpu-mi35x'
- 'nightly-8-gpu-mi35x-qwen3-235b-mxfp4'
- 'nightly-8-gpu-mi35x-qwen35'
- 'nightly-8-gpu-mi35x-kimi-k25'
- 'nightly-8-gpu-mi35x-glm5'
- 'nightly-8-gpu-mi35x-minimax-m25'
- 'nightly-accuracy-8-gpu-mi35x'
- 'nightly-8-gpu-mi35x-grok1-int4'
- 'nightly-8-gpu-mi35x-grok2'
- 'nightly-8-gpu-mi35x-deepseek-r1-mxfp4'
- 'nightly-accuracy-8-gpu-mi35x-deepseek-v32'
- 'nightly-accuracy-8-gpu-mi35x-deepseek-v32-mtp'
- 'nightly-perf-8-gpu-mi35x-deepseek-v32-basic'
- 'nightly-perf-8-gpu-mi35x-deepseek-v32-mtp'
workflow_call:
inputs:
ref:
@@ -98,7 +62,7 @@ jobs:
# ============================================== MI30x Unit Tests ==============================================
# 1-GPU Unit Tests - LoRA, debug utils, scheduler, etc. (MI30x only)
nightly-test-1-gpu-unit:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-test-1-gpu-unit')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-test-1-gpu-unit,'))
runs-on: linux-mi325-1gpu-sglang
steps:
- name: Checkout code
@@ -128,7 +92,7 @@ jobs:
# ============================================== MI30x Accuracy Tests ==============================================
# 2-GPU Accuracy Tests - GSM8K eval (MI30x only)
nightly-accuracy-2-gpu:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-accuracy-2-gpu')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-accuracy-2-gpu,'))
runs-on: linux-mi325-2gpu-sglang
steps:
- name: Checkout code
@@ -157,7 +121,7 @@ jobs:
# 2-GPU VLM Accuracy Tests - Vision-Language Models MMMU evaluation
nightly-accuracy-2-gpu-vlm:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-accuracy-2-gpu-vlm')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-accuracy-2-gpu-vlm,'))
runs-on: linux-mi325-2gpu-sglang
steps:
- name: Checkout code
@@ -187,7 +151,7 @@ jobs:
# 2-GPU Text Models Performance Tests
nightly-perf-2-gpu-text:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-perf-2-gpu-text')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-perf-2-gpu-text,'))
runs-on: linux-mi325-2gpu-sglang
steps:
- name: Checkout code
@@ -218,7 +182,7 @@ jobs:
# 2-GPU VLM Performance Tests
nightly-perf-2-gpu-vlm:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-perf-2-gpu-vlm')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-perf-2-gpu-vlm,'))
runs-on: linux-mi325-2gpu-sglang
steps:
- name: Checkout code
@@ -249,7 +213,7 @@ jobs:
# 8-GPU Accuracy Tests - GPT-OSS, Grok1-FP8 (accuracy only)
nightly-accuracy-8-gpu:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-accuracy-8-gpu')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-accuracy-8-gpu,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
@@ -289,7 +253,7 @@ jobs:
# ============================================== MI30x Combined Accuracy + Performance Tests ==============================================
# 8-GPU Grok1-INT4 (Accuracy + Performance combined)
nightly-8-gpu-grok1-int4:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-grok1-int4')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-grok1-int4,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
@@ -332,7 +296,7 @@ jobs:
# 8-GPU Grok2 (Accuracy + Performance combined)
nightly-8-gpu-grok2:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-grok2')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-grok2,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
@@ -375,7 +339,7 @@ jobs:
# 8-GPU DeepSeek-V3.1 (Accuracy + Performance combined)
nightly-8-gpu-deepseek-v31:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-deepseek-v31')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-deepseek-v31,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
@@ -418,7 +382,7 @@ jobs:
# 8-GPU DeepSeek-V3.2 (Basic Accuracy + Perf)
nightly-8-gpu-deepseek-v32:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-deepseek-v32')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-deepseek-v32,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
@@ -459,7 +423,7 @@ jobs:
# 8-GPU DeepSeek-V3.2 MTP (MTP Accuracy + Perf)
nightly-8-gpu-deepseek-v32-mtp:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-deepseek-v32-mtp')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-deepseek-v32-mtp,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
@@ -498,9 +462,39 @@ jobs:
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
exit ${TEST_EXIT_CODE:-0}
# 8-GPU DeepSeek-V3 KV FP8 (Basic + MTP with --kv-cache-dtype fp8_e4m3)
nightly-8-gpu-deepseek-v3-kv-fp8:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-deepseek-v3-kv-fp8,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
uses: actions/checkout@v4
with:
ref: ${{ inputs.ref || github.ref }}
- name: Setup docker
run: |
touch github_summary.md
bash scripts/ci/amd/amd_ci_start_container.sh
env:
GITHUB_WORKSPACE: ${{ github.workspace }}
- name: Install dependencies
run: bash scripts/ci/amd/amd_ci_install_dependency.sh
- name: DeepSeek-V3 KV FP8 Test (8-GPU Basic + MTP)
timeout-minutes: 120
run: |
> github_summary.md # Clear summary file
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
python3 run_suite.py --hw amd --suite nightly-amd-8-gpu-deepseek-v3-kv-fp8 --nightly --timeout-per-file 3600 ${{ inputs.continue_on_error && '--continue-on-error' || '' }} || TEST_EXIT_CODE=$?
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
exit ${TEST_EXIT_CODE:-0}
# 8-GPU Kimi-K2.5 (Accuracy)
nightly-8-gpu-kimi-k25:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-kimi-k25')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-kimi-k25,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
@@ -529,7 +523,7 @@ jobs:
exit ${TEST_EXIT_CODE:-0}
nightly-8-gpu-qwen3-235b:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-qwen3-235b')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-qwen3-235b,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
@@ -559,7 +553,7 @@ jobs:
# 8-GPU Qwen 3.5 (Accuracy)
nightly-8-gpu-qwen35:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-qwen35')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-qwen35,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
@@ -590,7 +584,7 @@ jobs:
exit ${TEST_EXIT_CODE:-0}
nightly-8-gpu-glm5:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-glm5')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-glm5,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
@@ -623,7 +617,7 @@ jobs:
# 8-GPU MiniMax-M2.5 (Accuracy)
nightly-8-gpu-minimax-m25:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-minimax-m25')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-minimax-m25,'))
runs-on: linux-mi325-8gpu-sglang
steps:
- name: Checkout code
@@ -655,7 +649,7 @@ jobs:
# ============================================== MI35x Tests ==============================================
# MI35x 1-GPU tests - platform-agnostic tests that may work on CDNA4 (gfx950)
nightly-test-1-gpu-mi35x:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-test-1-gpu-mi35x')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-test-1-gpu-mi35x,'))
runs-on: linux-mi35x-gpu-1
steps:
- name: Checkout code
@@ -687,7 +681,7 @@ jobs:
# MI35x 8-GPU Accuracy Tests - GPT-OSS (accuracy only)
nightly-accuracy-8-gpu-mi35x:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-accuracy-8-gpu-mi35x')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-accuracy-8-gpu-mi35x,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -719,7 +713,7 @@ jobs:
# MI35x 8-GPU Grok1-INT4 (Accuracy + Performance combined)
nightly-8-gpu-mi35x-grok1-int4:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-mi35x-grok1-int4')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-mi35x-grok1-int4,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -765,7 +759,7 @@ jobs:
# MI35x 8-GPU Grok2 (Accuracy + Performance combined)
nightly-8-gpu-mi35x-grok2:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-mi35x-grok2')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-mi35x-grok2,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -811,7 +805,7 @@ jobs:
# MI35x 8-GPU DeepSeek-R1-MXFP4 (Accuracy + Performance combined)
nightly-8-gpu-mi35x-deepseek-r1-mxfp4:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-mi35x-deepseek-r1-mxfp4')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-mi35x-deepseek-r1-mxfp4,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -853,9 +847,97 @@ jobs:
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
exit ${TEST_EXIT_CODE:-0}
# MI35x 8-GPU DeepSeek-R1-MXFP4 KV FP8 (Accuracy + Performance combined)
nightly-8-gpu-mi35x-deepseek-r1-mxfp4-kv-fp8:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-mi35x-deepseek-r1-mxfp4-kv-fp8,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
uses: actions/checkout@v4
with:
ref: ${{ inputs.ref || github.ref }}
- name: Setup docker
run: |
touch github_summary.md
bash scripts/ci/amd/amd_ci_start_container.sh
env:
GITHUB_WORKSPACE: ${{ github.workspace }}
- name: Install dependencies
run: |
bash scripts/ci/amd/amd_ci_install_dependency.sh
# Install tabulate for run_suite.py (missing in MI35x container)
bash scripts/ci/amd/amd_ci_exec.sh pip install tabulate
- name: Accuracy Test MI35x (8-GPU DeepSeek-R1-MXFP4 KV FP8)
timeout-minutes: 180
run: |
> github_summary.md # Clear summary file
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
python3 run_suite.py --hw amd --suite nightly-amd-8-gpu-mi35x-deepseek-r1-mxfp4-kv-fp8 --nightly --timeout-per-file 7200 ${{ inputs.continue_on_error && '--continue-on-error' || '' }} || TEST_EXIT_CODE=$?
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
exit ${TEST_EXIT_CODE:-0}
- name: Performance Test MI35x (8-GPU DeepSeek-R1-MXFP4 KV FP8)
timeout-minutes: 300
continue-on-error: true # Perf test failure doesn't fail the job if accuracy passed
run: |
> github_summary.md # Clear summary file
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
python3 registered/amd/perf/mi35x/test_deepseek_r1_mxfp4_kv_fp8_perf_mi35x.py || TEST_EXIT_CODE=$?
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
exit ${TEST_EXIT_CODE:-0}
# MI35x 8-GPU DeepSeek-R1-MXFP4 AllReduce Fusion (Accuracy + Performance combined)
nightly-8-gpu-mi35x-deepseek-r1-mxfp4-ar-fusion:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-mi35x-deepseek-r1-mxfp4-ar-fusion,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
uses: actions/checkout@v4
with:
ref: ${{ inputs.ref || github.ref }}
- name: Setup docker
run: |
touch github_summary.md
bash scripts/ci/amd/amd_ci_start_container.sh
env:
GITHUB_WORKSPACE: ${{ github.workspace }}
- name: Install dependencies
run: |
bash scripts/ci/amd/amd_ci_install_dependency.sh
# Install tabulate for run_suite.py (missing in MI35x container)
bash scripts/ci/amd/amd_ci_exec.sh pip install tabulate
- name: Accuracy Test MI35x (8-GPU DeepSeek-R1-MXFP4 AllReduce Fusion)
timeout-minutes: 180
run: |
> github_summary.md # Clear summary file
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
python3 run_suite.py --hw amd --suite nightly-amd-8-gpu-mi35x-deepseek-r1-mxfp4-ar-fusion --nightly --timeout-per-file 7200 ${{ inputs.continue_on_error && '--continue-on-error' || '' }} || TEST_EXIT_CODE=$?
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
exit ${TEST_EXIT_CODE:-0}
- name: Performance Test MI35x (8-GPU DeepSeek-R1-MXFP4 AllReduce Fusion)
timeout-minutes: 300
continue-on-error: true # Perf test failure doesn't fail the job if accuracy passed
run: |
> github_summary.md # Clear summary file
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
python3 registered/amd/perf/mi35x/test_deepseek_r1_mxfp4_ar_fusion_perf_mi35x.py || TEST_EXIT_CODE=$?
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
exit ${TEST_EXIT_CODE:-0}
# MI35x 8-GPU DeepSeek-V3.2 Accuracy Test
nightly-accuracy-8-gpu-mi35x-deepseek-v32:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-accuracy-8-gpu-mi35x-deepseek-v32')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-accuracy-8-gpu-mi35x-deepseek-v32,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -888,7 +970,7 @@ jobs:
# MI35x 8-GPU DeepSeek-V3.2 TP+MTP Accuracy Test
nightly-accuracy-8-gpu-mi35x-deepseek-v32-mtp:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-accuracy-8-gpu-mi35x-deepseek-v32-mtp')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-accuracy-8-gpu-mi35x-deepseek-v32-mtp,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -921,7 +1003,7 @@ jobs:
# MI35x 8-GPU DeepSeek-V3.2 Performance Test (Basic)
nightly-perf-8-gpu-mi35x-deepseek-v32-basic:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-perf-8-gpu-mi35x-deepseek-v32-basic')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-perf-8-gpu-mi35x-deepseek-v32-basic,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -954,7 +1036,7 @@ jobs:
# MI35x 8-GPU Kimi-K2.5 (Accuracy)
nightly-8-gpu-mi35x-kimi-k25:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-mi35x-kimi-k25')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-mi35x-kimi-k25,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -987,7 +1069,7 @@ jobs:
# MI35x 8-GPU Qwen3-235B-MXFP4 (Accuracy + Performance)
nightly-8-gpu-mi35x-qwen3-235b-mxfp4:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-mi35x-qwen3-235b-mxfp4')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-mi35x-qwen3-235b-mxfp4,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -1020,7 +1102,7 @@ jobs:
# MI35x 8-GPU Qwen 3.5 (Accuracy)
nightly-8-gpu-mi35x-qwen35:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-mi35x-qwen35')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-mi35x-qwen35,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -1052,7 +1134,7 @@ jobs:
exit ${TEST_EXIT_CODE:-0}
nightly-8-gpu-mi35x-glm5:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-mi35x-glm5')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-mi35x-glm5,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -1087,7 +1169,7 @@ jobs:
# MI35x 8-GPU MiniMax-M2.5 (Accuracy)
nightly-8-gpu-mi35x-minimax-m25:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-8-gpu-mi35x-minimax-m25')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-8-gpu-mi35x-minimax-m25,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -1120,7 +1202,7 @@ jobs:
# MI35x 8-GPU DeepSeek-V3.2 Performance Test (MTP)
nightly-perf-8-gpu-mi35x-deepseek-v32-mtp:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || inputs.job_filter == 'nightly-perf-8-gpu-mi35x-deepseek-v32-mtp')
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (inputs.job_filter == '' || inputs.job_filter == 'all' || contains(format(',{0},', inputs.job_filter), ',nightly-perf-8-gpu-mi35x-deepseek-v32-mtp,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -1169,6 +1251,7 @@ jobs:
- nightly-8-gpu-deepseek-v31
- nightly-8-gpu-deepseek-v32
- nightly-8-gpu-deepseek-v32-mtp
- nightly-8-gpu-deepseek-v3-kv-fp8
- nightly-8-gpu-kimi-k25
- nightly-8-gpu-qwen3-235b
- nightly-8-gpu-qwen35
@@ -1180,6 +1263,8 @@ jobs:
- nightly-8-gpu-mi35x-grok1-int4
- nightly-8-gpu-mi35x-grok2
- nightly-8-gpu-mi35x-deepseek-r1-mxfp4
- nightly-8-gpu-mi35x-deepseek-r1-mxfp4-kv-fp8
- nightly-8-gpu-mi35x-deepseek-r1-mxfp4-ar-fusion
- nightly-accuracy-8-gpu-mi35x-deepseek-v32
- nightly-accuracy-8-gpu-mi35x-deepseek-v32-mtp
- nightly-8-gpu-mi35x-kimi-k25

145
.github/workflows/pr-test-amd-rocm720.yml vendored
View File

@@ -28,7 +28,7 @@ on:
workflow_dispatch:
inputs:
target_stage:
description: "Specific stage to run (optional, for quick testing)"
description: "Specific stage(s) to run, comma-separated (e.g. 'stage-a-test-1-amd,stage-b-test-small-1-gpu-amd')"
required: false
type: string
default: ""
@@ -144,7 +144,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'sgl-kernel-unit-test-amd') ||
(contains(format(',{0},', inputs.target_stage), ',sgl-kernel-unit-test-amd,')) ||
(
!inputs.target_stage &&
needs.check-changes.outputs.sgl_kernel == 'true'
@@ -190,7 +190,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'sgl-kernel-unit-test-2-gpu-amd') ||
(contains(format(',{0},', inputs.target_stage), ',sgl-kernel-unit-test-2-gpu-amd,')) ||
(
!inputs.target_stage &&
needs.check-changes.outputs.sgl_kernel == 'true'
@@ -231,7 +231,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'stage-a-test-1-amd') ||
(contains(format(',{0},', inputs.target_stage), ',stage-a-test-1-amd,')) ||
(
!inputs.target_stage &&
(!failure() && !cancelled()) &&
@@ -270,7 +270,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'jit-kernel-unit-test-amd') ||
(contains(format(',{0},', inputs.target_stage), ',jit-kernel-unit-test-amd,')) ||
(
!inputs.target_stage &&
needs.check-changes.outputs.jit_kernel == 'true'
@@ -308,7 +308,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'stage-b-test-small-1-gpu-amd') ||
(contains(format(',{0},', inputs.target_stage), ',stage-b-test-small-1-gpu-amd,')) ||
(
!inputs.target_stage &&
(!failure() && !cancelled()) &&
@@ -319,7 +319,7 @@ jobs:
fail-fast: false
matrix:
runner: [linux-mi325-1gpu-sglang]
part: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12]
part: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13]
runs-on: ${{matrix.runner}}
steps:
- name: Checkout code
@@ -340,14 +340,14 @@ jobs:
- name: Run test
timeout-minutes: 30
run: |
bash scripts/ci/amd/amd_ci_exec.sh -w "/sglang-checkout/test" python3 run_suite.py --hw amd --suite stage-b-test-small-1-gpu-amd --auto-partition-id ${{ matrix.part }} --auto-partition-size 13 --timeout-per-file 1800 ${{ inputs.continue_on_error && '--continue-on-error' || '' }}
bash scripts/ci/amd/amd_ci_exec.sh -w "/sglang-checkout/test" python3 run_suite.py --hw amd --suite stage-b-test-small-1-gpu-amd --auto-partition-id ${{ matrix.part }} --auto-partition-size 14 --timeout-per-file 1800 ${{ inputs.continue_on_error && '--continue-on-error' || '' }}
stage-b-test-small-1-gpu-amd-nondeterministic:
needs: [check-changes]
if: |
always() &&
(
(inputs.target_stage == 'stage-b-test-small-1-gpu-amd-nondeterministic') ||
(contains(format(',{0},', inputs.target_stage), ',stage-b-test-small-1-gpu-amd-nondeterministic,')) ||
(
!inputs.target_stage &&
(!failure() && !cancelled()) &&
@@ -385,7 +385,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'stage-b-test-small-1-gpu-amd-mi35x') ||
(contains(format(',{0},', inputs.target_stage), ',stage-b-test-small-1-gpu-amd-mi35x,')) ||
(
!inputs.target_stage &&
(!failure() && !cancelled()) &&
@@ -423,7 +423,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'stage-b-test-large-1-gpu-amd') ||
(contains(format(',{0},', inputs.target_stage), ',stage-b-test-large-1-gpu-amd,')) ||
(
!inputs.target_stage &&
(!failure() && !cancelled()) &&
@@ -462,7 +462,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'stage-b-test-large-2-gpu-amd') ||
(contains(format(',{0},', inputs.target_stage), ',stage-b-test-large-2-gpu-amd,')) ||
(
!inputs.target_stage &&
(!failure() && !cancelled()) &&
@@ -501,7 +501,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'multimodal-gen-test-1-gpu-amd') ||
(contains(format(',{0},', inputs.target_stage), ',multimodal-gen-test-1-gpu-amd,')) ||
(
!inputs.target_stage &&
(!failure() && !cancelled()) &&
@@ -631,7 +631,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'multimodal-gen-test-2-gpu-amd') ||
(contains(format(',{0},', inputs.target_stage), ',multimodal-gen-test-2-gpu-amd,')) ||
(
!inputs.target_stage &&
(!failure() && !cancelled()) &&
@@ -760,7 +760,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'stage-c-test-large-8-gpu-amd') ||
(contains(format(',{0},', inputs.target_stage), ',stage-c-test-large-8-gpu-amd,')) ||
(
!inputs.target_stage &&
(!failure() && !cancelled()) &&
@@ -807,7 +807,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'stage-c-test-large-8-gpu-amd-mi35x') ||
(contains(format(',{0},', inputs.target_stage), ',stage-c-test-large-8-gpu-amd-mi35x,')) ||
(
!inputs.target_stage &&
(!failure() && !cancelled()) &&
@@ -841,6 +841,118 @@ jobs:
run: |
bash scripts/ci/amd/amd_ci_exec.sh -w "/sglang-checkout/test" python3 run_suite.py --hw amd --suite stage-c-test-large-8-gpu-amd-mi35x --auto-partition-id ${{ matrix.part }} --auto-partition-size 1 --timeout-per-file 3600 ${{ inputs.continue_on_error && '--continue-on-error' || '' }}
# =============================================== Disaggregation ====================================================
stage-b-test-large-8-gpu-35x-disaggregation-amd:
needs: [check-changes]
if: |
always() &&
(
(inputs.target_stage == 'stage-b-test-large-8-gpu-disaggregation-amd') ||
(
!inputs.target_stage &&
(!failure() && !cancelled()) &&
((needs.check-changes.outputs.main_package == 'true') || (needs.check-changes.outputs.sgl_kernel == 'true'))
)
)
strategy:
fail-fast: false
matrix:
runner: [linux-mi35x-gpu-8.fabric]
runs-on: ${{matrix.runner}}
steps:
- name: Checkout code
uses: actions/checkout@v4
with:
ref: ${{ inputs.pr_head_sha || inputs.ref || github.sha }}
- name: Ensure VRAM is clear
run: bash scripts/ensure_vram_clear.sh rocm
- name: Check Host RDMA Environment
id: rdma_detect
run: |
set +e
echo "=== Checking Host RDMA Environment ==="
echo ""
echo "=== 1. Ionic driver library check ==="
ls -l /usr/lib/x86_64-linux-gnu/libibverbs/libionic* 2>/dev/null || echo "libionic not found in standard path"
echo ""
echo "=== 2. Infiniband devices ==="
ls -la /dev/infiniband/ 2>/dev/null || echo "/dev/infiniband not found"
ls -la /sys/class/infiniband/ 2>/dev/null || echo "/sys/class/infiniband not found"
echo ""
echo "=== 3. ibv_devinfo ==="
which ibv_devinfo 2>/dev/null && ibv_devinfo 2>&1 || echo "ibv_devinfo not available"
echo ""
echo "=== 4. Kernel modules ==="
lsmod 2>/dev/null | grep -E "ib_|rdma|ionic" || echo "No RDMA kernel modules loaded"
echo ""
echo "=== 5. Detect RDMA Devices for test environment ==="
if [ -d "/sys/class/infiniband" ]; then
RDMA_DEVS=$(ls /sys/class/infiniband | paste -sd "," -)
echo "Detected RDMA Devices: $RDMA_DEVS"
echo "SGLANG_TEST_RDMA_DEVICE=$RDMA_DEVS" >> $GITHUB_ENV
else
echo "No RDMA devices found in /sys/class/infiniband"
echo "SGLANG_TEST_RDMA_DEVICE=" >> $GITHUB_ENV
fi
echo ""
echo "=== Host RDMA Check Complete ==="
- name: Start Special Container
run: bash scripts/ci/amd/amd_ci_start_container_disagg.sh --rocm-version rocm720
env:
GITHUB_WORKSPACE: ${{ github.workspace }}
- name: Install dependencies
run: bash scripts/ci/amd/amd_ci_install_dependency.sh
- name: Verify RDMA in Container
run: |
docker exec -u root ci_sglang bash -c '
echo "=== Container RDMA Verification ==="
echo "Device nodes:"
ls -la /dev/infiniband/
echo ""
echo "Provider libraries:"
ls /usr/lib/x86_64-linux-gnu/libibverbs/ | grep -E "ionic|mlx" || echo "No Ionic/Mellanox providers"
echo ""
echo "HCA devices:"
HCA_COUNT=$(ibv_devinfo -list 2>&1 | grep -oE "^[0-9]+ HCAs? found" | grep -oE "^[0-9]+" || echo "0")
ibv_devinfo -list
if [ "$HCA_COUNT" -gt 0 ]; then
echo ""
echo "=== SUCCESS: RDMA setup complete. Found $HCA_COUNT HCA(s) ==="
else
echo ""
echo "=== WARNING: No HCAs detected. RDMA tests may fail ==="
fi
'
- name: Run Aiter Op Test (RMSNorm)
timeout-minutes: 10
run: |
echo "Running pre-check: test_rmsnorm2d.py"
docker exec \
-e MAX_JOBS=192 \
ci_sglang \
python /sgl-workspace/aiter/op_tests/test_rmsnorm2d.py
- name: Run test_disaggregation
timeout-minutes: 60
run: |
bash scripts/ci/amd/amd_ci_exec.sh \
-e SGLANG_TEST_RDMA_DEVICE="${{ env.SGLANG_TEST_RDMA_DEVICE }}" \
-w "/sglang-checkout/test" python3 run_suite.py --hw amd --suite stage-b-test-large-8-gpu-35x-disaggregation-amd --timeout-per-file 1800 ${{ inputs.continue_on_error && '--continue-on-error' || '' }}
pr-test-amd-finish:
needs:
[
@@ -859,6 +971,7 @@ jobs:
stage-b-test-small-1-gpu-amd-mi35x,
stage-b-test-large-1-gpu-amd,
stage-b-test-large-2-gpu-amd,
stage-b-test-large-8-gpu-35x-disaggregation-amd,
stage-c-test-large-8-gpu-amd,
stage-c-test-large-8-gpu-amd-mi35x,
]

46
.github/workflows/pr-test-amd.yml vendored
View File

@@ -25,7 +25,7 @@ on:
workflow_dispatch:
inputs:
target_stage:
description: "Specific stage to run (optional, for quick testing)"
description: "Specific stage(s) to run, comma-separated (e.g. 'stage-a-test-1-amd,stage-b-test-small-1-gpu-amd')"
required: false
type: string
default: ""
@@ -141,7 +141,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'sgl-kernel-unit-test-amd') ||
(contains(format(',{0},', inputs.target_stage), ',sgl-kernel-unit-test-amd,')) ||
(
!inputs.target_stage &&
needs.check-changes.outputs.sgl_kernel == 'true'
@@ -188,7 +188,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'sgl-kernel-unit-test-2-gpu-amd') ||
(contains(format(',{0},', inputs.target_stage), ',sgl-kernel-unit-test-2-gpu-amd,')) ||
(
!inputs.target_stage &&
needs.check-changes.outputs.sgl_kernel == 'true'
@@ -230,7 +230,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'stage-a-test-1-amd') ||
(contains(format(',{0},', inputs.target_stage), ',stage-a-test-1-amd,')) ||
(
!inputs.target_stage &&
(!failure() && !cancelled()) &&
@@ -270,7 +270,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'jit-kernel-unit-test-amd') ||
(contains(format(',{0},', inputs.target_stage), ',jit-kernel-unit-test-amd,')) ||
(
!inputs.target_stage &&
needs.check-changes.outputs.jit_kernel == 'true'
@@ -309,7 +309,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'stage-b-test-small-1-gpu-amd') ||
(contains(format(',{0},', inputs.target_stage), ',stage-b-test-small-1-gpu-amd,')) ||
(
!inputs.target_stage &&
(!failure() && !cancelled()) &&
@@ -349,7 +349,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'stage-b-test-small-1-gpu-amd-nondeterministic') ||
(contains(format(',{0},', inputs.target_stage), ',stage-b-test-small-1-gpu-amd-nondeterministic,')) ||
(
!inputs.target_stage &&
(!failure() && !cancelled()) &&
@@ -388,7 +388,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'stage-b-test-small-1-gpu-amd-mi35x') ||
(contains(format(',{0},', inputs.target_stage), ',stage-b-test-small-1-gpu-amd-mi35x,')) ||
(
!inputs.target_stage &&
(!failure() && !cancelled()) &&
@@ -427,7 +427,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'stage-b-test-large-1-gpu-amd') ||
(contains(format(',{0},', inputs.target_stage), ',stage-b-test-large-1-gpu-amd,')) ||
(
!inputs.target_stage &&
(!failure() && !cancelled()) &&
@@ -467,7 +467,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'stage-b-test-large-2-gpu-amd') ||
(contains(format(',{0},', inputs.target_stage), ',stage-b-test-large-2-gpu-amd,')) ||
(
!inputs.target_stage &&
(!failure() && !cancelled()) &&
@@ -504,7 +504,15 @@ jobs:
multimodal-gen-test-1-gpu-amd:
needs: [check-changes]
if: needs.check-changes.outputs.multimodal_gen == 'true'
if: |
always() &&
(
(contains(format(',{0},', inputs.target_stage), ',multimodal-gen-test-1-gpu-amd,')) ||
(
!inputs.target_stage &&
needs.check-changes.outputs.multimodal_gen == 'true'
)
)
strategy:
fail-fast: false
max-parallel: 1 # Run one at a time to avoid eviction from resource exhaustion during AITER kernel JIT
@@ -624,7 +632,15 @@ jobs:
multimodal-gen-test-2-gpu-amd:
needs: [check-changes]
if: needs.check-changes.outputs.multimodal_gen == 'true'
if: |
always() &&
(
(contains(format(',{0},', inputs.target_stage), ',multimodal-gen-test-2-gpu-amd,')) ||
(
!inputs.target_stage &&
needs.check-changes.outputs.multimodal_gen == 'true'
)
)
strategy:
fail-fast: false
max-parallel: 1 # Run one at a time to avoid eviction from resource exhaustion during AITER kernel JIT
@@ -746,7 +762,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'stage-c-test-large-8-gpu-amd') ||
(contains(format(',{0},', inputs.target_stage), ',stage-c-test-large-8-gpu-amd,')) ||
(
!inputs.target_stage &&
(!failure() && !cancelled()) &&
@@ -794,7 +810,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'stage-c-test-large-8-gpu-amd-mi35x') ||
(contains(format(',{0},', inputs.target_stage), ',stage-c-test-large-8-gpu-amd-mi35x,')) ||
(
!inputs.target_stage &&
(!failure() && !cancelled()) &&
@@ -835,7 +851,7 @@ jobs:
if: |
always() &&
(
(inputs.target_stage == 'stage-b-test-large-8-gpu-disaggregation-amd') ||
(contains(format(',{0},', inputs.target_stage), ',stage-b-test-large-8-gpu-disaggregation-amd,')) ||
(
!inputs.target_stage &&
(!failure() && !cancelled()) &&