minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 13 Packages Projects Releases Wiki Activity

[AMD] Add Kimi-K3 8-GPU MI35x nightly accuracy CI (#32568)

Browse Source 
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 14:48:29 -07:00
committed by GitHub
co-authored by Cursor Agent Michael
parent 198a7b2fc9
commit 816ea65058
3 changed files with 185 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.github/workflows/nightly-test-amd-rocm720.yml
+21 -10
View File
@@ -71,9 +71,10 @@ on:
- nightly-8-gpu-mi35x-deepseek-v4-pro-rocm720
- nightly-8-gpu-mi35x-deepseek-v4-pro-mtp-rocm720
- nightly-8-gpu-mi35x-deepseek-v4-pro-dspark-rocm720
# 8-GPU Kimi-K2.6 (MI30x + MI35x)
# 8-GPU Kimi-K2.6 (MI30x)
- nightly-8-gpu-kimi-k26-rocm720
- nightly-8-gpu-mi35x-kimi-k26-rocm720
# 8-GPU Kimi-K3 (MI35x only - native MXFP4 needs gfx95x)
- nightly-8-gpu-mi35x-kimi-k3-rocm720
# 8-GPU Qwen3-235B (MI30x + MI35x MXFP4)
- nightly-8-gpu-qwen3-235b-rocm720
- nightly-8-gpu-mi35x-qwen3-235b-mxfp4-rocm720
@@ -1498,7 +1499,10 @@ jobs:
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
# ==============================================================================
# 8-GPU Kimi-K2.6 (MI30x + MI35x)
# 8-GPU Kimi-K2.6 (MI30x)
#
# Kept on MI30x: Kimi-K3 needs gfx95x for its native MXFP4 weights, so it
# cannot take over here the way it does on MI35x.
# ==============================================================================
nightly-8-gpu-kimi-k26-rocm720:
@@ -1534,8 +1538,12 @@ jobs:
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
exit ${TEST_EXIT_CODE:-0}
nightly-8-gpu-mi35x-kimi-k26-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-kimi-k26-rocm720,'))
# ==============================================================================
# 8-GPU Kimi-K3 (MI35x only - native MXFP4 needs gfx95x)
# ==============================================================================
nightly-8-gpu-mi35x-kimi-k3-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-kimi-k3-rocm720,'))
runs-on: linux-mi35x-gpu-8
steps:
- name: Checkout code
@@ -1560,13 +1568,15 @@ jobs:
# 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 Kimi-K2.6)
timeout-minutes: 180
# The 2.8T MXFP4 checkpoint is ~1.56 TB, so weight download and load
# dominate wall time well before the eval itself starts.
- name: Accuracy Test MI35x ROCm 7.2 (8-GPU Kimi-K3)
timeout-minutes: 300
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-accuracy-8-gpu-mi35x-kimi-k26 --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-accuracy-8-gpu-mi35x-kimi-k3 --nightly --timeout-per-file 14400 ${{ (github.event_name == 'schedule' || inputs.continue_on_error) && '--continue-on-error' || '' }} || TEST_EXIT_CODE=$?
echo "$(<github_summary.md )" >> $GITHUB_STEP_SUMMARY || true
exit ${TEST_EXIT_CODE:-0}
@@ -2137,9 +2147,10 @@ jobs:
- nightly-8-gpu-mi35x-deepseek-v4-pro-rocm720
- nightly-8-gpu-mi35x-deepseek-v4-pro-mtp-rocm720
- nightly-8-gpu-mi35x-deepseek-v4-pro-dspark-rocm720
# 8-GPU Kimi-K2.6 (MI30x + MI35x)
# 8-GPU Kimi-K2.6 (MI30x)
- nightly-8-gpu-kimi-k26-rocm720
- nightly-8-gpu-mi35x-kimi-k26-rocm720
# 8-GPU Kimi-K3 (MI35x only - native MXFP4 needs gfx95x)
- nightly-8-gpu-mi35x-kimi-k3-rocm720
# 8-GPU Qwen3-235B (MI30x + MI35x MXFP4)
- nightly-8-gpu-qwen3-235b-rocm720
- nightly-8-gpu-mi35x-qwen3-235b-mxfp4-rocm720