ascendai/triton

Triton compilation framework Triton-Ascend

Quick Reference

• Triton-Ascend is maintained byhttps://github.com/triton-lang/triton-ascend/
• Where to get help
  • https://github.com/triton-lang/triton-ascend/
  • https://triton-ascend.readthedocs.io/zh-cn/latest/index.html
  • https://github.com/triton-lang/triton-ascend/issues

---

Triton-Ascend

Triton-Ascend is a Triton compilation framework built for the Ascend platform, aiming to enable Triton code to run efficiently on Ascend hardware. For details, see https://github.com/triton-lang/triton-ascend/blob/main/README.md.

Supported Tags and Dockerfile

Tag specifications

Tag follow the following format：
{triton-ascend version}-{CANN version}-{torch_npu version}-{applicable product}-{os version}-{python version}

Triton-Ascend Images

Release 3.2.1

Key Image Components

Image List

Quick Start

Running the Triton-Ascend Container

bash
# Assume that your NPU device model is A3, the device is installed in /dev/davinci1, and the NPU driver is installed in /usr/local/Ascend:
docker run -u 0 -dit --shm-size=512g --name=triton-ascend_container --net=host --privileged \
--security-opt seccomp=unconfined \
--device=/dev/davinci0 \
--device=/dev/davinci1 \
--device=/dev/davinci2 \
--device=/dev/davinci3 \
--device=/dev/davinci4 \
--device=/dev/davinci5 \
--device=/dev/davinci6 \
--device=/dev/davinci7 \
--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/sbin/npu-smi:/usr/local/sbin/npu-smi \
-v /usr/local/Ascend/driver:/usr/local/Ascend/driver \
-v /etc/ascend_install.info:/etc/ascend_install.info \
-v /home:/home \
quay.io/ascend/triton:3.2.1-cann9.0.0-torch_npu2.7.1.post4-a3-ubuntu22.04-py3.11 \
/bin/bash

How to Build Locally?

arm64

bash
docker build \
--network host \
--build-arg TARGETPLATFORM=linux/arm64 \
-t triton:3.2.1-cann9.0.0-torch_npu2.7.1.post4-a3-ubuntun22.04-py3.11-aarch64 \
-f Dockerfile .

amd64

bash
docker build \
--network host \
--build-arg TARGETPLATFORM=linux/amd64 \
-t triton:3.2.1-cann9.0.0-torch_npu2.7.1.post4-a3-ubuntun22.04-py3.11-x86_64 \
-f Dockerfile .

How to Perform Secondary Development?

bash
# Using the triton-ascend image as the base image and adding user software to it.
FROM quay.io/ascend/triton:3.2.1-cann9.0.0-torch_npu2.7.1.post4-a3-ubuntu22.04-py3.11
RUN apt update -y && \
    apt install wget \
    ...

Supported hardware

License

View the license information of the CANN, Torch-npu and Triton-Ascend software contained in the image. For details, visit https://www.hiascend.com/en/software/protocol.
As with all container images, the pre-installed software packages (such as Python and system libraries) may be subject to their own licenses.