deepseek-ai/DeepEP-Ascend
- Total stars
- 258
- Today
- +2since 00:15 UTC
More actions
A high-performance communication library for machine learning training and inference on Huawei Ascend NPUs.
Save to My finds to follow releases and star growth. Saved in this browser.
About the project
A high-performance communication library for ML training and inference on Huawei Ascend NPUs: expert-parallel all-to-all for MoE dispatch/combine, plus PP, CP/DP and Engram. Public buffer APIs are aligned with NVIDIA DeepEP.
Useful for
- Run MoE training with expert-parallel dispatch/combine on an Ascend NPU cluster
- Speed up MoE inference with FP8 dispatch and deferred epilogues
- Port code from NVIDIA DeepEP to Ascend without changing the buffer API
README summarized by DeepSeek V4.1 Flash. Details may be inaccurate. Read the original README →
How to get started
What you need
- Linux on an Ascend host with supported NPU drivers and firmware.
- Ascend 950 (A5) with UBMEM connectivity and UBCCTP/URMA channels between the participating ranks.
- An even number of ranks for multi-rank communication.
- CANN with Ascend C, the Bisheng compiler, HCCL and HCOMM headers/libraries.
- Python 3.10 or newer, a matching PyTorch/torchnpu installation, and NumPy for the tests.
- A C++20 compiler and standard library supporting std::format.
git clone https://github.com/deepseek-ai/DeepEP-Ascend.git
cd DeepEP-Ascend
# Fetch the required DeepJIT submodule over HTTPS.
git submodule update --init --recursive third-party/deep_jit
python -m pip install --no-build-isolation .
Copied as is from the “quick start” section of the README. GitNova has not run these commands: check them in the original before running. Full instructions →
Star growth looks organic. Data as of October 16, 2026, 23:43 UTC.
The star-growth assessment does not verify whether the project is safe to run.
Stars per day
Bars show stars per UTC day; the line is a moving average of available history. Gaps mean missing data, not zero growth. Today's bar is incomplete. Red marks spike days.
Why it’s trending
- Star-counter measurements on 2026-10-16 (UTC), 00:15–23:40: 256 → 258 stars (+2). This is the change over that interval.
- The repository is 16 days old and already has 258 stars.
- Recent forks include notable developers: @gfreewind (268 followers).
Measured growth: the last 7 completed days
The full 7-day window is unavailable; no weekly total or comparison is calculated.
Daily measurements · UTC
- November 9, 2026
- —
- November 10, 2026
- —
- November 11, 2026
- —
- November 12, 2026
- —
- November 13, 2026
- —
- November 14, 2026
- —
- November 15, 2026
- —
Completed days come from GitHub star history; today and forecasts are excluded from this comparison. History checked: October 16, 2026, 23:43 UTC. Sources and measurement limits →
Three GitHub discoveries in each edition
What they do, why they are gaining interest, and what to check before using them.
Editions at 10:00, 16:00, 22:00 MSK.
Numbers
- Total stars
- 258
- Today
- +2 stars measured 2026-10-16, 00:15–23:40 UTC
- Forks
- 32
- Issues and pull requests
- 5
- Watchers
- 1
- Language
- C++
- Created
- September 30, 2026
- Last push
- September 30, 2026
Star trust
Growth looks organic: forks and discussion are in line with active projects, and stars arrive unevenly, the way people give them.
These are heuristics, not a verdict: we judge by the repository’s behavior, not by a list of stargazers.
Spotted in
- October 6, 2026Top new repositories this week: #57
- October 5, 2026Top new repositories this week: #61
- October 4, 2026Top new repositories this week: #42
- October 3, 2026Top new repositories this week: #48
- October 2, 2026Top new repositories this week: #57
- October 1, 2026Top new repositories this week: #61
- September 30, 2026Top new repositories this week: #82