Skip to content

[AMD] add unit-test-backend-8-gpu-amd back#15253

Merged
HaiShaw merged 9 commits intosgl-project:mainfrom
yctseng0211:enable_dpsk
Dec 19, 2025
Merged

[AMD] add unit-test-backend-8-gpu-amd back#15253
HaiShaw merged 9 commits intosgl-project:mainfrom
yctseng0211:enable_dpsk

Conversation

@yctseng0211
Copy link
Copy Markdown
Collaborator

@yctseng0211 yctseng0211 commented Dec 16, 2025 •
edited
Loading

Motivation

(#14383) fixed the regression of deepseek_v3, so adds the related test case back.

Modifications

For test_deepseek_v3_mtp.py, the avg_spec_accept_length in test_a_gsm8k is 2.84 ~ 2.88, and 2.864 in upstream ci, so we relax the threshold avg_spec_accept_length from 2.9 to 2.8 if is_in_amd_ci(), as the implementation in test_bs_1_speed
Change the runner from linux-mi300-gpu-8 to linux-mi325-gpu-8

Accuracy Tests

Device: mi300

Test Case: test_deepseek_v3_basic.py

SGLANG_IS_IN_CI_AMD=1 SGLANG_IS_IN_CI=1 python3 run_suite.py --suite per-commit-8-gpu-amd --auto-partition-id 0 --auto-partition-size 2 --timeout-per-file 3600

Accuracy: 0.946
Invalid: 0.000
Latency: 40.063 s
Output throughput: 3556.277 token/s
metrics={'accuracy': 0.9461713419257013, 'invalid': 0.0, 'latency': 40.06267931126058, 'output_throughput': 3556.277374587731}

Test Case: test_deepseek_v3_mtp.py

SGLANG_IS_IN_CI_AMD=1 SGLANG_IS_IN_CI=1 python3 run_suite.py --suite per-commit-8-gpu-amd --auto-partition-id 1 --auto-partition-size 2 --timeout-per-file 3600

Accuracy: 0.960
Invalid: 0.000
Latency: 40.102 s
Output throughput: 510.873 token/s
metrics={'accuracy': 0.96, 'invalid': 0.0, 'latency': 40.101929197087884, 'output_throughput': 510.87317767963447}
avg_spec_accept_length=2.8482719865130655

Benchmarking and Profiling

Checklist

@gemini-code-assist
Copy link
Copy Markdown
Contributor

gemini-code-assist Bot commented Dec 16, 2025

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 16, 2025
@yctseng0211 yctseng0211 marked this pull request as ready for review December 17, 2025 03:31
HaiShaw
HaiShaw approved these changes Dec 19, 2025
View reviewed changes
Copy link
Copy Markdown
Collaborator

@HaiShaw HaiShaw left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM - changes apply to amd only.

@HaiShaw HaiShaw merged commit af780c5 into sgl-project:main Dec 19, 2025
90 of 95 checks passed
Prozac614 pushed a commit to Prozac614/sglang that referenced this pull request Dec 23, 2025
@yctseng0211 @saienduri @Prozac614
[AMD] add unit-test-backend-8-gpu-amd back (sgl-project#15253)
9d60413 
Co-authored-by: Sai Enduri <saimanas.enduri@amd.com>
jiaming1130 pushed a commit to zhuyijie88/sglang that referenced this pull request Dec 25, 2025
@yctseng0211 @saienduri @jiaming1130
[AMD] add unit-test-backend-8-gpu-amd back (sgl-project#15253)
f0e7954 
Co-authored-by: Sai Enduri <saimanas.enduri@amd.com>
YChange01 pushed a commit to YChange01/sglang that referenced this pull request Jan 13, 2026
@yctseng0211 @saienduri @YChange01
[AMD] add unit-test-backend-8-gpu-amd back (sgl-project#15253)
d2766d4 
Co-authored-by: Sai Enduri <saimanas.enduri@amd.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@HaiShaw HaiShaw HaiShaw approved these changes

@merrymercy merrymercy Awaiting requested review from merrymercy merrymercy is a code owner

@Fridge003 Fridge003 Awaiting requested review from Fridge003 Fridge003 is a code owner

@ispobock ispobock Awaiting requested review from ispobock ispobock is a code owner

@Kangyan-Zhou Kangyan-Zhou Awaiting requested review from Kangyan-Zhou Kangyan-Zhou is a code owner

Assignees

No one assigned

Labels

amd deepseek run-ci

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@yctseng0211 @HaiShaw @saienduri