mirror of
https://github.com/deepseek-ai/DeepEP.git
synced 2026-11-05 00:47:32 +00:00
[GH-ISSUE #396] Why the internode_ll::dispatch and internode_ll::combine use all SMs? #268
Labels
No labels
pull-request
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
deepseek-ai/DeepEP#268
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @ican2002 on GitHub (Sep 8, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/396
@LyricZhao , hello, I read the code for low latency:
github.com/deepseek-ai/DeepEP@c18eabdebf/csrc/deep_ep.cpp (L1169-L1182)github.com/deepseek-ai/DeepEP@c18eabdebf/csrc/deep_ep.cpp (L1274-L1285)I found that num_device_sms (the number of SMs) is from,
github.com/deepseek-ai/DeepEP@c18eabdebf/csrc/deep_ep.cpp (L43-L46)Does it means that all SMs are used for dispatch and combine? to minimize the time cost?
As all SMs are used, will this affect the compute-communication overlap? or we should assume that there is no computation as Experts are waiting for dispatch/combind?
Thanks.
@sphish commented on GitHub (Sep 10, 2025):
You can get answer from https://github.com/deepseek-ai/profile-data?tab=readme-ov-file#decoding