πŸ’« KLX xpu

[δΈ­ζ–‡η‰ˆ|English]

πŸ’« KLX xpu#

  • Based on Triton 3.0, x64

  • Available for P800

1. Build and run environment#

1.1 Use the image (P800)#

If your network connection is available, you do not need to perform the later step 1.x, because dependencies will be fetched automatically during the build.

# Plan A: docker pull (44.9GB)
IMAGE=harbor.baai.ac.cn/flagtree/flagtree-xpu-py310-torch2.5.1-ubuntu20.04:202604-base
docker pull ${IMAGE}
# Plan B: docker load (21GB)
IMAGE=flagtree-xpu-py310-torch2.5.1-ubuntu20.04:202604-base
wget https://baai-cp-web.ks3-cn-beijing.ksyuncs.com/trans/flagtree-xpu-py310-torch2.5.1-ubuntu20.04.202604-base.tar.gz
docker load -i flagtree-xpu-py310-torch2.5.1-ubuntu20.04.202604-base.tar.gz

CONTAINER=flagtree-dev-xxx
docker run -dit \
    --net=host --privileged \
    --ulimit stack=67108864 --ulimit memlock=-1 \
    --ulimit nofile=120000 --shm-size=256g \
    --group-add video --cap-add=SYS_PTRACE --cap-add=SYS_ADMIN \
    --security-opt seccomp=unconfined \
    --device=/dev/xpu0:/dev/xpu0 --device=/dev/xpu1:/dev/xpu1 \
    --device=/dev/xpu2:/dev/xpu2 --device=/dev/xpu3:/dev/xpu3 \
    --device=/dev/xpu4:/dev/xpu4 --device=/dev/xpu5:/dev/xpu5 \
    --device=/dev/xpu6:/dev/xpu6 --device=/dev/xpu7:/dev/xpu7 \
    --device=/dev/xpuctrl:/dev/xpuctrl --device /dev/fuse \
    -v /etc/localtime:/etc/localtime:ro \
    -v /data:/data -v /home:/home -v /tmp:/tmp \
    -w /root --name ${CONTAINER} ${IMAGE} bash
docker exec -it ${CONTAINER} /bin/bash

1.2 Manually download the FlagTree dependencies#

mkdir -p ~/.flagtree/xpu; cd ~/.flagtree/xpu
wget https://klx-sdk-release-public.su.bcebos.com/v1/triton/flaggems/2025_4_season/llvm/20260304/XTDK-llvm19-ubuntu2004_x86_64.tar.gz
tar zxvf XTDK-llvm19-ubuntu2004_x86_64.tar.gz
wget https://baai-cp-web.ks3-cn-beijing.ksyuncs.com/trans/xre-Linux-x86_64_v0.3.0.tar.gz
tar zxvf xre-Linux-x86_64_v0.3.0.tar.gz
wget https://klx-sdk-release-public.su.bcebos.com/XTriton/xpu-device-libs-ubuntu-x64_v0.3.6.1.1.tar.gz
tar zxvf xpu-device-libs-ubuntu-x64_v0.3.6.1.1.tar.gz

1.3 Manually download the Triton dependencies#

The Triton dependencies are already downloaded and installed in the image. If you do not need to build FlagTree or Triton from source, you do not need to download the Triton dependencies.

cd ${YOUR_CODE_DIR}/FlagTree
# For Triton 3.1 (x64)
wget https://baai-cp-web.ks3-cn-beijing.ksyuncs.com/trans/build-deps-triton_3.1.x-linux-x64.tar.gz
sh python/scripts/unpack_triton_build_deps.sh ./build-deps-triton_3.1.x-linux-x64.tar.gz

After executing the above script, the original ~/.triton directory will be renamed, and a new ~/.triton directory will be created to store the pre-downloaded packages. Note that the script will prompt for manual confirmation during execution.

2. Installation Commands#

2.1 Source-free Installation#

# Note: First install PyTorch, then execute the following commands
python3 -m pip uninstall -y triton  # Repeat the cmd until fully uninstalled
RES="--index-url=https://resource.flagos.net/repository/flagos-pypi-hosted/simple"
python3.10 -m pip install flagtree===0.5.1+xpu3.0 $RES

After installing flagtree, you can check it with:

python3 -m pip show flagtree

2.2 Build from Source#

cd ${YOUR_CODE_DIR}/FlagTree/python
export FLAGTREE_BACKEND=xpu
MAX_JOBS=32 python3 -m pip install . --no-build-isolation -v

3. Testing and validation#

Before testing, you need to execute export XPU_EVENT_KL3_ENABLE=1

Refer to Tests of xpu backend