deepseek-ai/DeepEP
1
0
Fork 0
mirror of https://github.com/deepseek-ai/DeepEP.git synced 2026-11-05 00:47:32 +00:00
Code Issues 308 Projects Packages Wiki Activity

[PR #448] [MERGED] Decouple the dispatch and buffer allocation on hybrid-ep #577

New issue
Closed
opened 2026-10-27 21:04:13 +00:00 by bsakaguchi · 0 comments
bsakaguchi commented 2026-10-27 21:04:13 +00:00
Owner
Copy link

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepEP/pull/448
Author: @Autumn1998
Created: 10/10/2025
Status: ✅ Merged
Merged: 10/15/2025
Merged by: @jershi425

Base: hybrid-ep ← Head: tongliu/fix_bug

📝 Commits (9)

  • 2e92368 fix bug if num_dispatched_tokens = 0
  • 4ef04b2 fix bug on open handlr for scale
  • 41b6445 use None as default
  • 017a520 runtime config
  • 6a9548c finish runtime config
  • 3ae80eb set boundary of the max_seq_len
  • 4c33543 Merge pull request #1 from Autumn1998/runtime_config
  • 3f5876c bitwise equeal on dispatch
  • cd018bd rename Ep toEP

📊 Changes

11 files changed (+445 additions, -279 deletions)

View changed files

📝 Hybrid-EP_Intra-node_Implementation.md (+1 -1)
📝 csrc/hybrid_ep/config.cuh (+25 -0)
📝 csrc/hybrid_ep/executor/executor.cu (+16 -9)
📝 csrc/hybrid_ep/executor/executor.cuh (+1 -1)
📝 csrc/hybrid_ep/hybrid_ep.cu (+121 -101)
📝 csrc/hybrid_ep/hybrid_ep.cuh (+22 -19)
📝 csrc/hybrid_ep/pybind_hybrid_ep.cu (+37 -9)
📝 csrc/hybrid_ep/utils.cuh (+38 -0)
📝 deep_ep/__init__.py (+1 -1)
📝 deep_ep/hybrid_ep_buffer.py (+159 -120)
📝 tests/test_hybrid_ep.py (+24 -18)

📄 Description

  1. Decouple the dispatch/combine with buffer allocation, if dispatch required bigger space than the pre-allocated buffer, we will free the old buffers and allocate new.
  2. fix bug if no tokens received on the local rank.
  3. fix bug on trying to open handle from scale tensor on bf16
  4. Use None as default value instead of -1

🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.

## 📋 Pull Request Information **Original PR:** https://github.com/deepseek-ai/DeepEP/pull/448 **Author:** [@Autumn1998](https://github.com/Autumn1998) **Created:** 10/10/2025 **Status:** ✅ Merged **Merged:** 10/15/2025 **Merged by:** [@jershi425](https://github.com/jershi425) **Base:** `hybrid-ep` ← **Head:** `tongliu/fix_bug` --- ### 📝 Commits (9) - [`2e92368`](https://github.com/deepseek-ai/DeepEP/commit/2e9236852b6646697e1910bce210ec2adad9bca8) fix bug if num_dispatched_tokens = 0 - [`4ef04b2`](https://github.com/deepseek-ai/DeepEP/commit/4ef04b2e0b515c9219e94276988ab1aae9e7120d) fix bug on open handlr for scale - [`41b6445`](https://github.com/deepseek-ai/DeepEP/commit/41b6445cd364e1df25c3173ffe150d20b950e129) use None as default - [`017a520`](https://github.com/deepseek-ai/DeepEP/commit/017a520224046dcf347e13ed56f1aacd8bbba89e) runtime config - [`6a9548c`](https://github.com/deepseek-ai/DeepEP/commit/6a9548c204a6e0e746983b59905b725e264d13f2) finish runtime config - [`3ae80eb`](https://github.com/deepseek-ai/DeepEP/commit/3ae80ebd9035e905257b8caf1b32b07370b31108) set boundary of the max_seq_len - [`4c33543`](https://github.com/deepseek-ai/DeepEP/commit/4c3354312e148421542e16637654880e8c26785b) Merge pull request #1 from Autumn1998/runtime_config - [`3f5876c`](https://github.com/deepseek-ai/DeepEP/commit/3f5876c7960e849f33f55d8598461227772fe35a) bitwise equeal on dispatch - [`cd018bd`](https://github.com/deepseek-ai/DeepEP/commit/cd018bd62ae0f52f6b5452d8f15225e72d03838f) rename Ep toEP ### 📊 Changes **11 files changed** (+445 additions, -279 deletions) <details> <summary>View changed files</summary> 📝 `Hybrid-EP_Intra-node_Implementation.md` (+1 -1) 📝 `csrc/hybrid_ep/config.cuh` (+25 -0) 📝 `csrc/hybrid_ep/executor/executor.cu` (+16 -9) 📝 `csrc/hybrid_ep/executor/executor.cuh` (+1 -1) 📝 `csrc/hybrid_ep/hybrid_ep.cu` (+121 -101) 📝 `csrc/hybrid_ep/hybrid_ep.cuh` (+22 -19) 📝 `csrc/hybrid_ep/pybind_hybrid_ep.cu` (+37 -9) 📝 `csrc/hybrid_ep/utils.cuh` (+38 -0) 📝 `deep_ep/__init__.py` (+1 -1) 📝 `deep_ep/hybrid_ep_buffer.py` (+159 -120) 📝 `tests/test_hybrid_ep.py` (+24 -18) </details> ### 📄 Description 1. Decouple the dispatch/combine with buffer allocation, if dispatch required bigger space than the pre-allocated buffer, we will free the old buffers and allocate new. 2. fix bug if no tokens received on the local rank. 3. fix bug on trying to open handle from scale tensor on bf16 4. Use None as default value instead of -1 --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>
bsakaguchi 2026-10-27 21:04:13 +00:00
Sign in to join this conversation.
No Branch/Tag specified
Branches Tags
main
hybrid-ep
chore/remove-clang-format-yapf
v2.5
nvDev
try-fix-cu128
mori-ep
antgroup-opt
compat-old-torch
add-linter
eager-rdma
v1.2.1
Labels
Clear labels   
pull-request
Mirrored from GitHub Pull Request

No labels
pull-request
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
deepseek-ai/DeepEP#577
No description provided.