mirror of
https://github.com/BillyOutlast/flash-attention-prebuild-wheels-rocm.git
synced 2026-07-01 01:37:53 -04:00
Junya Morioka
225 lines
7.4 KiB
YAML
225 lines
7.4 KiB
YAML
name: Test workflow
|
|
|
|
on:
|
|
workflow_dispatch:
|
|
|
|
jobs:
|
|
# #########################################################
|
|
# Build wheels with GitHub hosted runner
|
|
# #########################################################
|
|
build_wheels:
|
|
name: Build wheels
|
|
runs-on: ubuntu-22.04
|
|
env:
|
|
DEBIAN_FRONTEND: noninteractive
|
|
TERM: xterm-256color
|
|
timeout-minutes: 1000
|
|
strategy:
|
|
fail-fast: false
|
|
matrix:
|
|
flash-attn-version: ["2.4.3", "2.5.9", "2.6.3"]
|
|
python-version: ["3.10", "3.11", "3.12"]
|
|
torch-version: ["2.8.0.dev20250523"]
|
|
# # https://developer.nvidia.com/cuda-toolkit-archive
|
|
cuda-version: ["12.8.1"]
|
|
exclude:
|
|
# torch < 2.2 does not support Python 3.12
|
|
- python-version: "3.12"
|
|
torch-version: "2.0.1"
|
|
- python-version: "3.12"
|
|
torch-version: "2.1.2"
|
|
# torch 2.0.1 does not support CUDA 12.x
|
|
- torch-version: "2.0.1"
|
|
cuda-version: "12.1.1"
|
|
- torch-version: "2.0.1"
|
|
cuda-version: "12.4.1"
|
|
- torch-version: "2.0.1"
|
|
cuda-version: "12.6.3"
|
|
- torch-version: "2.0.1"
|
|
cuda-version: "12.8.1"
|
|
# torch 2.7.0 does not support CUDA 12.4
|
|
- torch-version: "2.7.0"
|
|
cuda-version: "12.4.1"
|
|
steps:
|
|
- uses: actions/checkout@v4
|
|
|
|
- name: Maximize build space
|
|
run: |
|
|
df -h
|
|
echo "-----------------------------"
|
|
sudo rm -rf /usr/share/dotnet
|
|
sudo rm -rf /usr/local/lib/android
|
|
sudo rm -rf /opt/ghc
|
|
sudo rm -rf /opt/hostedtoolcache/CodeQL
|
|
df -h
|
|
|
|
- name: Set Swap Space
|
|
uses: pierotofy/set-swap-space@master
|
|
with:
|
|
swap-size-gb: 48
|
|
|
|
- uses: actions/setup-python@v5
|
|
with:
|
|
python-version: ${{ matrix.python-version }}
|
|
|
|
- uses: Jimver/cuda-toolkit@master
|
|
with:
|
|
cuda: ${{ matrix.cuda-version }}
|
|
sub-packages: '["nvcc", "toolkit"]'
|
|
method: "network"
|
|
|
|
- name: Install build dependencies
|
|
run: |
|
|
sudo apt install -y ninja-build clang
|
|
pip install -U pip setuptools==75.8.0 wheel setuptools packaging psutil
|
|
|
|
- name: Set environment variables
|
|
run: |
|
|
export PATH=/usr/local/nvidia/bin:/usr/local/nvidia/lib64:$PATH
|
|
export LD_LIBRARY_PATH=/usr/local/nvidia/lib64:/usr/local/cuda/lib64:$LD_LIBRARY_PATH
|
|
export MAX_JOBS=2
|
|
export NVCC_THREADS=2
|
|
export FLASH_ATTENTION_FORCE_BUILD=TRUE
|
|
|
|
- name: Build wheels
|
|
timeout-minutes: 800
|
|
run: |
|
|
chmod +x build_linux.sh
|
|
./build_linux.sh ${{ matrix.flash-attn-version }} ${{ matrix.python-version }} ${{ matrix.torch-version }} ${{ matrix.cuda-version }}
|
|
wheel_name=$(basename $(ls flash-attention/dist/*.whl | head -n 1))
|
|
echo "wheel_name=$wheel_name" >> $GITHUB_ENV
|
|
|
|
- name: Install Test
|
|
run: |
|
|
pip install flash-attention/dist/${{ env.wheel_name }}
|
|
python -c "import flash_attn; print(flash_attn.__version__)"
|
|
|
|
# #########################################################
|
|
# Build wheels with self-hosted runner
|
|
# #########################################################
|
|
# build_wheels_self_hosted:
|
|
# name: Build wheels
|
|
# runs-on: self-hosted
|
|
# container:
|
|
# image: ubuntu:22.04
|
|
# defaults:
|
|
# run:
|
|
# shell: bash
|
|
# env:
|
|
# DEBIAN_FRONTEND: noninteractive
|
|
# TERM: xterm-256color
|
|
# timeout-minutes: 1000
|
|
# strategy:
|
|
# fail-fast: false
|
|
# matrix:
|
|
# flash-attn-version: ["2.7.4"]
|
|
# python-version: ["3.10", "3.11", "3.12"]
|
|
# torch-version: ["2.8.0.dev20250523"]
|
|
# # https://developer.nvidia.com/cuda-toolkit-archive
|
|
# cuda-version: ["12.8.1"]
|
|
# exclude:
|
|
# # torch < 2.2 does not support Python 3.12
|
|
# - python-version: "3.12"
|
|
# torch-version: "2.0.1"
|
|
# - python-version: "3.12"
|
|
# torch-version: "2.1.2"
|
|
# # torch 2.0.1 does not support CUDA 12.x
|
|
# - torch-version: "2.0.1"
|
|
# cuda-version: "12.1.1"
|
|
# - torch-version: "2.0.1"
|
|
# cuda-version: "12.4.1"
|
|
# - torch-version: "2.0.1"
|
|
# cuda-version: "12.6.3"
|
|
# - torch-version: "2.0.1"
|
|
# cuda-version: "12.8.1"
|
|
# # torch 2.7.0 does not support CUDA 12.4
|
|
# - torch-version: "2.7.0"
|
|
# cuda-version: "12.4.1"
|
|
|
|
# steps:
|
|
# - name: Install tools
|
|
# shell: bash
|
|
# run: |
|
|
# apt-get update && apt-get install -y --no-install-recommends \
|
|
# curl \
|
|
# ca-certificates \
|
|
# sudo \
|
|
# software-properties-common \
|
|
# wget \
|
|
# unzip \
|
|
# zip \
|
|
# git \
|
|
# build-essential \
|
|
# gcc \
|
|
# g++ \
|
|
# clang \
|
|
# ninja-build \
|
|
# keyboard-configuration
|
|
|
|
# - uses: actions/checkout@v4
|
|
|
|
# - uses: actions/setup-python@v5
|
|
# with:
|
|
# python-version: ${{ matrix.python-version }}
|
|
|
|
# - uses: Jimver/cuda-toolkit@master
|
|
# env:
|
|
# DEBIAN_FRONTEND: noninteractive
|
|
# with:
|
|
# cuda: ${{ matrix.cuda-version }}
|
|
# sub-packages: '["nvcc", "toolkit"]'
|
|
# method: "network"
|
|
|
|
# - name: Install build dependencies
|
|
# run: |
|
|
# pip install -U pip setuptools==75.8.0 wheel setuptools packaging psutil
|
|
|
|
# - name: Set environment variables
|
|
# run: |
|
|
# export PATH=/usr/local/nvidia/bin:/usr/local/nvidia/lib64:$PATH
|
|
# export LD_LIBRARY_PATH=/usr/local/nvidia/lib64:/usr/local/cuda/lib64:$LD_LIBRARY_PATH
|
|
# export MAX_JOBS=2
|
|
# export NVCC_THREADS=2
|
|
# export FLASH_ATTENTION_FORCE_BUILD=TRUE
|
|
|
|
# - name: Build wheels
|
|
# timeout-minutes: 800
|
|
# run: |
|
|
# chmod +x build_linux.sh
|
|
# ./build_linux.sh ${{ matrix.flash-attn-version }} ${{ matrix.python-version }} ${{ matrix.torch-version }} ${{ matrix.cuda-version }}
|
|
# wheel_name=$(basename $(ls flash-attention/dist/*.whl | head -n 1))
|
|
# echo "wheel_name=$wheel_name" >> $GITHUB_ENV
|
|
|
|
# - name: Install Test
|
|
# run: |
|
|
# pip install flash-attention/dist/${{ env.wheel_name }}
|
|
# python -c "import flash_attn; print(flash_attn.__version__)"
|
|
|
|
# - name: Get the tag version
|
|
# id: extract_branch
|
|
# shell: bash
|
|
# run: echo "branch=${GITHUB_REF#refs/tags/}" >> $GITHUB_OUTPUT
|
|
|
|
# - name: Get Release with Tag
|
|
# id: get_release
|
|
# uses: joutvhu/get-release@v1
|
|
# env:
|
|
# GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
|
|
# with:
|
|
# tag_name: ${{ steps.extract_branch.outputs.branch }}
|
|
|
|
# - name: Upload Release Asset
|
|
# uses: actions/upload-release-asset@v1
|
|
# env:
|
|
# GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
|
|
# with:
|
|
# upload_url: ${{ steps.get_release.outputs.upload_url }}
|
|
# asset_path: flash-attention/dist/${{ env.wheel_name }}
|
|
# asset_name: ${{ env.wheel_name }}
|
|
# asset_content_type: application/*
|
|
|
|
# - name: Clean up
|
|
# if: always()
|
|
# run: |
|
|
# sudo rm -rf /opt/hostedtoolcache/Python/${{ matrix.python-version }}*
|