Skip to content

[AMD] Fix and add accuracy-test-2-gpu-amd back#15415

Merged
HaiShaw merged 4 commits intosgl-project:mainfrom
yctseng0211:fix_eval_moe
Dec 19, 2025
Merged

[AMD] Fix and add accuracy-test-2-gpu-amd back#15415
HaiShaw merged 4 commits intosgl-project:mainfrom
yctseng0211:fix_eval_moe

Conversation

@yctseng0211
Copy link
Copy Markdown
Collaborator

@yctseng0211 yctseng0211 commented Dec 18, 2025

Co-author : @bingxche

Motivation

To add accuracy-test-2-gpu-amd(test_moe_eval_accuracy_large.py) back, we add SGLANG_USE_AITER_AR=0 into the command, disabling the aiter custom_allreduce in this test.

Modifications

.github/workflows/pr-test-amd.yml

Accuracy Tests

Device - MI-325
Settings - SGLANG_USE_AITER_AR=0, num_threads=1024
Before - 0.268
After - 0.422
Ran 3 tests in 406.697s

Score: **0.422**
Writing report to /tmp/humaneval_mistralai_Mixtral-8x7B-Instruct-v0.1.html
{'pass@1': 0.42195121951219516, 'pass@1:std': 0.47524689096990586, 'pass@2': 0.4445121951219513, 'pass@2:std': 0.4832881593555032, 'pass@5': 0.4695121951219512, 'pass@5:std': 0.499069628161961, 'score:std': 0.47524689096990586, 'score': 0.42195121951219516}
[2025-12-18 17:51:32 TP0] Decode batch, #running-req: 2, #token: 3282, token usage: 0.00, cuda graph: True, gen throughput (token/s): 204.54, #queue-req: 0, 
[2025-12-18 17:51:32 TP0] Decode batch, #running-req: 1, #token: 1733, token usage: 0.00, cuda graph: True, gen throughput (token/s): 169.94, #queue-req: 0, 
[2025-12-18 17:51:32 TP0] Decode batch, #running-req: 1, #token: 1773, token usage: 0.00, cuda graph: True, gen throughput (token/s): 123.76, #queue-req: 0, 
100%|█████████████████████████████████████████████████████████████████████████████████████████| 5000/5000 [02:59<00:00, 27.92it/s]
Total latency: 179.171 s
Score: 0.633
Writing report to /tmp/mmlu_mistralai_Mixtral-8x7B-Instruct-v0.1.html
{'other': 0.7285223367697594, 'other:std': 0.4447218699336571, 'score:std': 0.4820967122891422, 'stem': 0.6374321880650995, 'stem:std': 0.48074150401607635, 'humanities': 0.5173237753882916, 'humanities:std': 0.49969979668426523, 'social_sciences': 0.7045454545454546, 'social_sciences:std': 0.45624681590647115, 'score': 0.6326}
Writing results to /tmp/mmlu_mistralai_Mixtral-8x7B-Instruct-v0.1.json
WARNING:root:GITHUB_STEP_SUMMARY environment variable not set
.
----------------------------------------------------------------------
Ran 3 tests in 406.697s

OK

Settings - SGLANG_USE_AITER_AR=1(default), num_threads=1
Before - 0.268
After - 0.421
Ran 3 tests in 1096.271s

Total latency: 848.003 s
Score: **0.421**
Writing report to /tmp/humaneval_mistralai_Mixtral-8x7B-Instruct-v0.1.html
{'pass@1': 0.42073170731707316, 'pass@1:std': 0.4867106630046756, 'pass@2': 0.4292682926829269, 'pass@2:std': 0.493491247081309, 'pass@5': 0.4329268292682927, 'pass@5:std': 0.49548076629471227, 'score:std': 0.4867106630046756, 'score': 0.42073170731707316}
Writing results to /tmp/humaneval_mistralai_Mixtral-8x7B-Instruct-v0.1.json
WARNING:root:GITHUB_STEP_SUMMARY environment variable not set
Total latency: 183.573 s
Score: 0.640
Writing report to /tmp/mmlu_mistralai_Mixtral-8x7B-Instruct-v0.1.html
{'other': 0.7285223367697594, 'other:std': 0.4447218699336571, 'score:std': 0.4801164858656699, 'stem': 0.6482820976491862, 'stem:std': 0.4775064601832704, 'humanities': 0.5286738351254481, 'humanities:std': 0.4991771340708613, 'social_sciences': 0.7083333333333334, 'social_sciences:std': 0.45452967144315476, 'score': 0.6396}
Writing results to /tmp/mmlu_mistralai_Mixtral-8x7B-Instruct-v0.1.json
WARNING:root:GITHUB_STEP_SUMMARY environment variable not set
.
----------------------------------------------------------------------
Ran 3 tests in 1096.271s

OK

Benchmarking and Profiling

Checklist

@gemini-code-assist
Copy link
Copy Markdown
Contributor

Note

Gemini is unable to generate a summary for this pull request due to the file types involved not being currently supported.

@github-actions github-actions Bot added the amd label Dec 18, 2025
@yctseng0211 yctseng0211 marked this pull request as ready for review December 19, 2025 03:18
@HaiShaw HaiShaw merged commit a21aa87 into sgl-project:main Dec 19, 2025
83 of 85 checks passed
Prozac614 pushed a commit to Prozac614/sglang that referenced this pull request Dec 23, 2025
Co-authored-by: bingxche <Bingxu.Chen@amd.com>
jiaming1130 pushed a commit to zhuyijie88/sglang that referenced this pull request Dec 25, 2025
Co-authored-by: bingxche <Bingxu.Chen@amd.com>
YChange01 pushed a commit to YChange01/sglang that referenced this pull request Jan 13, 2026
Co-authored-by: bingxche <Bingxu.Chen@amd.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants