[AMD] Add GLM-5.2 MI35x nightly accuracy and perf benchmark (#32570)

This commit is contained in:
Michael
2026-08-19 22:49:43 -07:00
committed by GitHub
parent 58e327480a
commit 02b93e7e01
4 changed files with 231 additions and 10 deletions
+21 -10
View File
@@ -82,9 +82,10 @@ on:
- nightly-8-gpu-qwen35-rocm720
- nightly-8-gpu-mi35x-qwen35-rocm720
- nightly-8-gpu-mi35x-qwen35-triton-dcp-rocm720
# 8-GPU GLM-5.1 (MI30x + MI35x)
# 8-GPU GLM-5.1 (MI30x)
- nightly-8-gpu-glm51-rocm720
- nightly-8-gpu-mi35x-glm51-rocm720
# 8-GPU GLM-5.2-FP8 (MI35x accuracy + performance)
- nightly-8-gpu-mi35x-glm52-fp8-rocm720
# 8-GPU GLM-5-MXFP4 (MI35x only)
- nightly-8-gpu-mi35x-glm5-mxfp4-rocm720
# 4-GPU MiniMax-M2.5 (MI35x)
@@ -1852,8 +1853,12 @@ jobs:
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
exit ${TEST_EXIT_CODE:-0}
nightly-8-gpu-mi35x-glm51-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (!(inputs.job_filter || inputs.job_select) || (inputs.job_filter || inputs.job_select) == 'all' || contains(format(',{0},', inputs.job_filter || inputs.job_select), ',nightly-8-gpu-mi35x-glm51-rocm720,'))
# ==============================================================================
# 8-GPU GLM-5.2-FP8 (MI35x accuracy + performance)
# ==============================================================================
nightly-8-gpu-mi35x-glm52-fp8-rocm720:
if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') && (!(inputs.job_filter || inputs.job_select) || (inputs.job_filter || inputs.job_select) == 'all' || contains(format(',{0},', inputs.job_filter || inputs.job_select), ',nightly-8-gpu-mi35x-glm52-fp8-rocm720,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -1870,30 +1875,35 @@ jobs:
bash scripts/ci/amd/amd_ci_start_container.sh --rocm-version rocm720
env:
GITHUB_WORKSPACE: ${{ github.workspace }}
ENABLE_CACHE_HOST: "1"
- name: Install dependencies
run: |
bash scripts/ci/amd/amd_ci_install_dependency.sh --skip-test-time-deps
bash scripts/ci/amd/amd_ci_exec.sh pip install tabulate
- name: Accuracy Test MI35x ROCm 7.2 (8-GPU GLM-5.1 DSA)
# The ~700 GB FP8 checkpoint means weight download and load dominate wall
# time well before the eval itself starts.
- name: Accuracy Test MI35x ROCm 7.2 (8-GPU GLM-5.2-FP8 DSA)
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-glm51 --nightly --timeout-per-file 7200 ${{ (github.event_name == 'schedule' || inputs.continue_on_error) && '--continue-on-error' || '' }} || TEST_EXIT_CODE=$?
python3 run_suite.py --hw amd --suite nightly-amd-8-gpu-mi35x-glm52-fp8 --nightly --timeout-per-file 7200 ${{ (github.event_name == 'schedule' || inputs.continue_on_error) && '--continue-on-error' || '' }} || TEST_EXIT_CODE=$?
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
exit ${TEST_EXIT_CODE:-0}
- name: Performance Test MI35x ROCm 7.2 (8-GPU GLM-5.1)
# Run perf only after correctness passes. Keeping both steps in this job
# reuses the cached ~700 GB checkpoint and one scarce 8-GPU MI35x slot.
- name: Performance Test MI35x ROCm 7.2 (8-GPU GLM-5.2-FP8)
timeout-minutes: 120
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 run_suite.py --hw amd --suite nightly-perf-8-gpu-mi35x-glm51 --nightly --timeout-per-file 5400 ${{ (github.event_name == 'schedule' || inputs.continue_on_error) && '--continue-on-error' || '' }} || TEST_EXIT_CODE=$?
python3 run_suite.py --hw amd --suite nightly-perf-8-gpu-mi35x-glm52-fp8 --nightly --timeout-per-file 5400 ${{ (github.event_name == 'schedule' || inputs.continue_on_error) && '--continue-on-error' || '' }} || TEST_EXIT_CODE=$?
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
exit ${TEST_EXIT_CODE:-0}
@@ -2175,9 +2185,10 @@ jobs:
- nightly-8-gpu-qwen35-rocm720
- nightly-8-gpu-mi35x-qwen35-rocm720
- nightly-8-gpu-mi35x-qwen35-triton-dcp-rocm720
# 8-GPU GLM-5.1 (MI30x + MI35x)
# 8-GPU GLM-5.1 (MI30x)
- nightly-8-gpu-glm51-rocm720
- nightly-8-gpu-mi35x-glm51-rocm720
# 8-GPU GLM-5.2-FP8 (MI35x accuracy + performance)
- nightly-8-gpu-mi35x-glm52-fp8-rocm720
# 8-GPU GLM-5-MXFP4 (MI35x only)
- nightly-8-gpu-mi35x-glm5-mxfp4-rocm720
# 4-GPU MiniMax-M2.5 (MI35x)