Skip to content

Modify the description of DeepEP in the README file. #348

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

Merged
Yael-X merged 1 commit into from
Jan 30, 2026
Merged

Modify the description of DeepEP in the README file. #348

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 2 additions & 3 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,15 +25,14 @@ For contribution guidelines, please refer to the [Contribution Guide](docs/devel
DeepEP-Ascend provides optimized all-to-all communication kernels for Expert Parallelism in MoE models.

**Communication Modes:**
- **Normal Mode**: High-throughput dispatch and combine operations for training and prefill phases (up to 4096 tokens/batch)
- **Normal Mode**: High-throughput dispatch and combine operations for training and prefill phases (up to 65536 tokens/batch for A3 and 8192 tokens/batch for A2)
- **Low-Latency Mode**: Optimized for production inference with small batch sizes (128 tokens/batch), achieving sub-150us latency

**Key Capabilities:**
- Token dispatch and combine with automatic load balancing
- Fused MoE computation (`fused_deep_moe`)
- Intranode HCCS and internode RDMA communication
- A3 full-mesh HCCS communication and A2 Intranode HCCS + internode RDMA communication
- INT8/FP8/BF16 quantization for reduced memory bandwidth
- Support for EP scales: 2, 4, 8, 16, 32, 64, 128, 144, 160 ranks

### SGLang-Kernel-NPU

Expand Down