vllm-ascend

| About Ascend | Documentation | Support Matrix | #SIG-Ascend | Users Forum | Weekly Meeting |

English | 中文

Latest News 🔥

• [2026/09] We released the new release candidate v0.26.0rc1! Please follow the official guide to start using vLLM Ascend Plugin on Ascend.
• [2026/08] We released the new official version v0.23.0! Please follow the official guide to start using vLLM Ascend Plugin on Ascend.
• [2026/05] We released the new official version v0.18.0! Please follow the official guide to start using vLLM Ascend Plugin on Ascend.
• [2026/02] We released the new official version v0.13.0! Please follow the official guide to start using vLLM Ascend Plugin on Ascend.

vLLM Ascend (vllm-ascend) is a community maintained hardware plugin for running vLLM seamlessly on the Huawei Ascend NPU.

It is the recommended approach for supporting the Ascend backend within the vLLM community. It adheres to the principles outlined in the [RFC]: Hardware pluggable, providing a hardware-pluggable interface that decouples the integration of the Ascend NPU with vLLM.

By using vLLM Ascend plugin, popular open-source models, including Transformer-like, Mixture-of-Experts (MoE), Embedding, Multi-modal LLMs can run seamlessly on the Ascend NPU.

For detailed information on supported models and features, please refer to the support matrix.

DeepWiki is a dynamic knowledge base collaboratively maintained by the community and AI, designed to provide you with deeper technical insights beyond conventional documentation. If official documentation serves as a "quick start" guide to help you get going, the DeepWiki is your technical companion for "deep understanding". Here, you can explore: core architecture and design principles, key source code interpretations, technical context, and decision-making logic.

Whether you are a performance tuner looking to deploy vLLM-Ascend in production or a contributor aiming to build upon it for secondary development, DeepWiki offers invaluable references for you. Welcome to explore now and join us in delving deep into the technical core of vLLM-Ascend.

• Hardware: Atlas 800I A2 Inference series, Atlas A2 Training series, Atlas 800I A3 Inference series, Atlas A3 Training series, Atlas 300I Duo (Experimental)
• OS: Linux
• Software:
  • Python >= 3.10, < 3.13
  • CANN == 9.1.0 (For Ascend HDK version, please refer to the Release Notes)
  • PyTorch == 2.10.0, TorchNPU == 2.10.0.post4
  • vLLM (the same version as vllm-ascend)

If you need to access Ascend NPU computing resources for development or testing, please visit the HiDevLab - Online Development page on the HiDevLab platform to apply for and use them.

Please use the following recommended versions to get started quickly:

vllm-ascend has a main branch and a dev branch.

• main: main branch, corresponds to the vLLM main branch, and is continuously monitored for quality through Ascend CI.
• releases/vX.Y.Z: development branch, created alongside new releases of vLLM. For example, releases/v0.13.0 is the dev branch for vLLM v0.13.0 version.

Below are the maintained branches:

Please refer to Versioning policy for more details.

See CONTRIBUTING for more details, which is a step-by-step guide to help you set up the development environment, build and test.

We welcome and value any contributions and collaborations:

• Please let us know if you encounter a bug by filing an issue
• Please use User forum for usage questions and help.

• vLLM Ascend Weekly Meeting: https://tinyurl.com/vllm-ascend-meeting
• Wednesday, 15:00 - 16:00 (UTC+8, Convert to your timezone)

Apache License 2.0, as found in the LICENSE file.