openeuler/aibrix repository overview

⁠Quick reference

• The official AIBrix Ascend docker images

• Maintained by: openEuler CloudNative SIG⁠

• Where to get help: openEuler CloudNative SIG⁠, openEuler⁠

⁠AIBrix | openEuler

Current AIBrix docker images are built on the openEuler⁠. This repository is free to use and exempted from per-user rate limits.

AIBrix⁠ is an open-source cloud-native inference infrastructure platform built around vLLM. It provides a high-performance inference gateway, routing, token billing, observability and elastic autoscaling on Kubernetes, and also offers a local mode that runs the same gateway stack on a single machine without Kubernetes.

AIBrix is a Kubernetes platform project whose quick start routes requests from an Envoy-fronted Go gateway to a Python inference backend. This image packages that Python/inference component for Ascend NPUs: vLLM with vLLM-Ascend⁠ (vllm==0.23.0, vllm-ascend==0.23.0) on top of torch_npu⁠, together with the AIBrix release source tree at /opt/aibrix for the local mode deployment scripts. The Go gateway and Envoy are installed separately following the upstream quick start. The software stack is validated on a single Ascend 910B4 card with CANN 9.1.0, torch==2.10.0, torch_npu==2.10.0.post4, vllm==0.23.0 and vllm-ascend==0.23.0 on Python 3.12, following the Ascend quick start guard⁠.

Read more about Ascend at hiascend.com⁠ and explore the AIBrix documentation at aibrix.readthedocs.io⁠.

⁠Supported tags and respective Dockerfile links

The tag of each AIBrix docker image consists of the AIBrix version and the version of the software stack. The details are as follows

⁠Usage

⁠Quick start 1: supported devices

• Atlas A2 Training series (Atlas 800T A2, Atlas 900 A2 PoD, Atlas 200T A2 Box16, Atlas 300T A2)

• Atlas 800I A2 Inference series (Atlas 800I A2)

⁠Quick start 2: setup environment using container

# Update DEVICE according to your device (/dev/davinci[0-7])
export DEVICE=/dev/davinci0
# Update the aibrix image
export IMAGE=openeuler/aibrix:0.8.0-torch_npu2.10.0-cann9.1.0-python3.12-oe2403sp4
docker run --rm \
--name my-aibrix \
--device $DEVICE \
--device /dev/davinci_manager \
--device /dev/devmm_svm \
--device /dev/hisi_hdc \
-v /usr/local/dcmi:/usr/local/dcmi \
-v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \
-v /usr/local/Ascend/driver/lib64/:/usr/local/Ascend/driver/lib64/ \
-v /usr/local/Ascend/driver/version.info:/usr/local/Ascend/driver/version.info \
-v /etc/ascend_install.info:/etc/ascend_install.info \
-v /root/.cache:/root/.cache \
-it $IMAGE bash

⁠Quick start 3: run the vLLM-Ascend backend for AIBrix local mode

Validate the environment inside the container:

python3 -c "import torch, torch_npu; print('torch=', torch.__version__); print('torch_npu=', torch_npu.__version__); print('is_available:', torch.npu.is_available())"
python3 -c "import importlib.metadata as m; print('vllm', m.version('vllm')); print('vllm-ascend', m.version('vllm-ascend'))"

Start the OpenAI-compatible inference backend that AIBrix local mode routes to (uses the public Qwen/Qwen2.5-0.5B-Instruct model, downloaded from ModelScope, no Hugging Face token required):

export VLLM_USE_MODELSCOPE=True
vllm serve Qwen/Qwen2.5-0.5B-Instruct \
  --served-model-name Qwen/Qwen2.5-0.5B-Instruct \
  --host 127.0.0.1 \
  --port 8000 \
  --max-model-len 2048 \
  --max-num-seqs 4 \
  --gpu-memory-utilization 0.2

Once the server is up, the vLLM log must show backend=hccl, which confirms the Ascend communication backend is active. Send a chat completion request:

curl -fsS http://127.0.0.1:8000/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{"model":"Qwen/Qwen2.5-0.5B-Instruct","messages":[{"role":"user","content":"Say hi in one sentence."}],"max_tokens":32,"temperature":0}'

To run the full AIBrix local mode on top of this backend, build the Go gateway from the source tree bundled at /opt/aibrix and install Envoy, then start deployment/local/run-local.sh with an endpoints.yaml pointing at the backend, following the AIBrix local mode quick start⁠. The gateway then serves the same API on port 10080.

For faster model download in China, use the ModelScope or hf-mirror channel:

export HF_ENDPOINT=https://hf-mirror.com
# or
pip install modelscope

⁠Question and answering

If you have any questions or want to use some special features, please submit an issue or a pull request on openeuler-docker-images⁠.