Thanks to visit codestin.com
Credit goes to github.com

Skip to content

Opt moe block by dlblas, when ep > 1 #3461

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Draft
wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

Opt moe block by dlblas, when ep > 1 #3461

wants to merge 1 commit into from

Conversation

hellozmz
Copy link

@hellozmz hellozmz commented Apr 21, 2025 •
edited
Loading

4节点测试数据：

Input token throughput (tok/s):

Output len base dlblas 性能变化
1 9638.95 11468.43 +19.0%
32 12263.80 13783.17 +12.4%
64 12641.08 13994.80 +10.7%
128 12512.95 14468.12 +15.6%
512 10759.34 12104.70 +12.5%
1k 8355.66 9178.18 +9.8%
2k 5965.40 6327.07 +6.06%
4k 3711.70 3892.58 +4.9%
8k 1959.26 1937.02 -1.1%
16k 997.58 1012.90 +1.5%
32k 443.81 436.11 -1.7%

Output token throughput (tok/s):

Output len base dlblas 性能变化
1 4.70 5.59 +18.9%
32 192.23 216.04 +12.4%
64 400.91 443.84 +10.7%
128 793.26 917.21 +15.6%
512 2733.50 3075.30 +12.5%
1k 4118.29 4523.69 +9.8%
2k 5918.32 6277.14 +6.06%
4k 7287.26 7642.39 +4.87%
8k 7956.68 7866.38 -1.13%
16k 7796.37 7916.10 +1.54%
32k 7048.98 6926.75 -1.73%

ref: DeepLink-org/dlBLAS#24

@hellozmz hellozmz force-pushed the zmz/lmdeploy_opt_by_dlblas branch from 7e7d32a to bcb9807 Compare April 21, 2025 03:43
@hellozmz hellozmz changed the title when ep > 1, opt moe block by dlblas Opt moe block by dlblas, when ep > 1 Apr 21, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@hellozmz @zhaochaoxing