[AMD] Add the Kimi-K3 MI35x perf benchmarks in nightly (#34985)

Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Michael <michaelzhang-ai@users.noreply.github.com>
This commit is contained in:
Michael
2026-08-17 21:23:51 -07:00
committed by GitHub
co-authored by Cursor Agent Michael
parent d528192bf9
commit 8ea5229d42
2 changed files with 155 additions and 0 deletions
@@ -1580,6 +1580,23 @@ jobs:
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
exit ${TEST_EXIT_CODE:-0}
# Shares the job with the eval above so the checkpoint is already cached
# and only one 8-GPU MI35x slot is consumed. Step ordering is also the
# accuracy gate: a failed eval fails the job before this runs, so
# throughput is never measured on a build that got the tokens wrong.
# continue-on-error matches every other combined accuracy + perf job here,
# so a perf hiccup cannot turn the accuracy result red.
- name: Performance Test MI35x ROCm 7.2 (8-GPU Kimi-K3)
timeout-minutes: 300
continue-on-error: true
run: |
> github_summary.md # Clear summary file
bash scripts/ci/amd/amd_ci_exec.sh -w /sglang-checkout/test \
-e GITHUB_STEP_SUMMARY="/sglang-checkout/github_summary.md" \
python3 registered/amd/perf/mi35x/test_kimi_k3_perf_mi35x.py || TEST_EXIT_CODE=$?
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
exit ${TEST_EXIT_CODE:-0}
# ==============================================================================
# 8-GPU Qwen3-235B (MI30x + MI35x MXFP4)
# ==============================================================================