mirror of
https://github.com/deepseek-ai/DeepEP.git
synced 2026-11-05 00:47:32 +00:00
[GH-ISSUE #344] NVL/RDMA tuning process (auto tune) #238
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#238
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 @dshirron3 on GitHub (Jul 30, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/344
Regarding the tuning process mentioned in README:
under: Auto-tuning on your cluster:
For better performance on your cluster, we recommend to run all the tests and use the best auto-tuned configuration. The default configurations are optimized on the DeepSeek's internal cluster
Can you elaborate on tuning process?
for example, if i get a certain optimal nvl chunk size and rdma chunk size, should that go to into buffer.py -> get_dispatch/combine_config
Currently the number are for example:
Config(Buffer.num_sms, 24, 256, 6, 128)
What is each number here, and what is the relation to the nvl chunk numbers i see in test_internode.py results
Example from test_internode.py:
[tuning] Best dispatch (BF16): SMs 24, NVL chunk 12, RDMA chunk 32: 42.36 GB/s (RDMA), 138.25 GB/s (NVL)
@sphish commented on GitHub (Aug 5, 2025):
You can refer to this.
github.com/deepseek-ai/DeepEP@26cf250a33/tests/test_internode.py (L186)