mirror of
https://github.com/BillyOutlast/flash-attention-prebuild-wheels-rocm.git
synced 2026-07-01 01:27:54 -04:00
Junya Morioka
394 lines
13 KiB
YAML
394 lines
13 KiB
YAML
name: Build wheels and upload to GitHub Releases
|
|
|
|
on:
|
|
push:
|
|
tags:
|
|
- "v*"
|
|
|
|
jobs:
|
|
create_releases:
|
|
name: Create Releases
|
|
runs-on: ubuntu-latest
|
|
steps:
|
|
- uses: actions/checkout@v4
|
|
- name: Create Release
|
|
env:
|
|
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
|
|
run: |
|
|
gh release create "${{ github.ref_name }}" \
|
|
--title "${{ github.ref_name }}" \
|
|
--notes "## Linux x86_64
|
|
|
|
| Flash-Attention | Python | PyTorch | CUDA |
|
|
| --- | --- | --- | --- |
|
|
| 2.8.0 | 3.10, 3.11, 3.12 | 2.4.1,2.5.1,2.6.0,2.7.1 | 12.1.1,12.4.1,12.8.1 |"
|
|
|
|
# ## Windows x86_64
|
|
|
|
# | Flash-Attention | Python | PyTorch | CUDA |
|
|
# | --- | --- | --- | --- |
|
|
# | 2.5.9, 2.6.3, 2.7.4 | 3.10, 3.11, 3.12 | 2.4.1, 2.5.1, 2.6.0, 2.7.1 | 12.4.1, 12.8.1 |"
|
|
|
|
# #########################################################
|
|
# Build wheels with GitHub hosted runner
|
|
# #########################################################
|
|
build_wheels:
|
|
name: Build wheels and Upload (Linux x86_64, GitHub hosted runner)
|
|
needs: create_releases
|
|
runs-on: ubuntu-22.04
|
|
env:
|
|
DEBIAN_FRONTEND: noninteractive
|
|
TERM: xterm-256color
|
|
MAX_JOBS: 2
|
|
NVCC_THREADS: 2
|
|
strategy:
|
|
fail-fast: false
|
|
matrix:
|
|
flash-attn-version: ["2.8.0"]
|
|
python-version: ["3.10", "3.11", "3.12"]
|
|
torch-version: ["2.4.1","2.5.1","2.6.0"]
|
|
# # https://developer.nvidia.com/cuda-toolkit-archive
|
|
cuda-version: ["12.1.1","12.4.1"]
|
|
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 2.6.0 does not support CUDA 12.1
|
|
- torch-version: "2.6.0"
|
|
cuda-version: "12.1.1"
|
|
# 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: ${{ matrix.cuda-version }}
|
|
sub-packages: '["nvcc", "toolkit"]'
|
|
method: "network"
|
|
use-github-cache: false
|
|
use-local-cache: false
|
|
|
|
- name: Install build dependencies
|
|
run: |
|
|
sudo apt install -y ninja-build clang
|
|
pip install -U pip setuptools==75.8.0 wheel setuptools packaging psutil
|
|
|
|
- name: Set environment variables
|
|
run: |
|
|
export PATH=/usr/local/nvidia/bin:/usr/local/nvidia/lib64:$PATH
|
|
export LD_LIBRARY_PATH=/usr/local/nvidia/lib64:/usr/local/cuda/lib64:$LD_LIBRARY_PATH
|
|
|
|
- name: Build wheels
|
|
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: Upload Release Asset
|
|
env:
|
|
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
|
|
run: |
|
|
tag_name=${{ github.ref_name }}
|
|
wheel_path="flash-attention/dist/${{ env.wheel_name }}"
|
|
|
|
# Check if the file exists
|
|
if [ ! -f "$wheel_path" ]; then
|
|
echo "Error: Wheel file not found at $wheel_path"
|
|
exit 1
|
|
fi
|
|
|
|
# Upload the release asset using GitHub CLI
|
|
gh release upload "$tag_name" "$wheel_path" --clobber
|
|
|
|
echo "Successfully uploaded $wheel_name to release $tag_name"
|
|
|
|
# #########################################################
|
|
# Build wheels with self-hosted runner
|
|
# #########################################################
|
|
build_wheels_self_hosted:
|
|
name: Build wheels and Upload (Linux x86_64, self-hosted runner)
|
|
needs: create_releases
|
|
runs-on: self-hosted
|
|
container:
|
|
image: ubuntu:22.04
|
|
defaults:
|
|
run:
|
|
shell: bash
|
|
env:
|
|
DEBIAN_FRONTEND: noninteractive
|
|
TERM: xterm-256color
|
|
MAX_JOBS: 2
|
|
NVCC_THREADS: 2
|
|
timeout-minutes: 1000
|
|
strategy:
|
|
fail-fast: false
|
|
matrix:
|
|
flash-attn-version: ["2.8.0"]
|
|
python-version: ["3.10", "3.11", "3.12"]
|
|
torch-version: ["2.4.1","2.5.1","2.6.0","2.7.1"]
|
|
# https://developer.nvidia.com/cuda-toolkit-archive
|
|
cuda-version: ["12.8.1"]
|
|
exclude:
|
|
# torch < 2.2 does not support Python 3.12
|
|
- python-version: "3.12"
|
|
torch-version: "2.0.1"
|
|
- python-version: "3.12"
|
|
torch-version: "2.1.2"
|
|
# torch 2.0.1 does not support CUDA 12.x
|
|
- torch-version: "2.0.1"
|
|
cuda-version: "12.1.1"
|
|
- torch-version: "2.0.1"
|
|
cuda-version: "12.4.1"
|
|
- torch-version: "2.0.1"
|
|
cuda-version: "12.6.3"
|
|
- torch-version: "2.0.1"
|
|
cuda-version: "12.8.1"
|
|
# torch 2.6.0 does not support CUDA 12.1
|
|
- torch-version: "2.6.0"
|
|
cuda-version: "12.1.1"
|
|
# torch 2.7.0 does not support CUDA 12.4
|
|
- torch-version: "2.7.0"
|
|
cuda-version: "12.4.1"
|
|
steps:
|
|
- name: Install tools
|
|
shell: bash
|
|
run: |
|
|
apt-get update && apt-get install -y --no-install-recommends \
|
|
curl \
|
|
ca-certificates \
|
|
sudo \
|
|
software-properties-common \
|
|
wget \
|
|
unzip \
|
|
zip \
|
|
git \
|
|
build-essential \
|
|
gcc \
|
|
g++ \
|
|
clang \
|
|
ninja-build \
|
|
keyboard-configuration
|
|
|
|
- name: Install gh
|
|
shell: bash
|
|
run: |
|
|
sudo mkdir -p -m 755 /etc/apt/keyrings
|
|
out=$(mktemp)
|
|
wget -nv -O$out https://cli.github.com/packages/githubcli-archive-keyring.gpg
|
|
cat $out | sudo tee /etc/apt/keyrings/githubcli-archive-keyring.gpg > /dev/null
|
|
sudo chmod go+r /etc/apt/keyrings/githubcli-archive-keyring.gpg
|
|
echo "deb [arch=$(dpkg --print-architecture) signed-by=/etc/apt/keyrings/githubcli-archive-keyring.gpg] https://cli.github.com/packages stable main" | sudo tee /etc/apt/sources.list.d/github-cli.list > /dev/null
|
|
sudo apt update
|
|
sudo apt install gh -y
|
|
|
|
- uses: actions/checkout@v4
|
|
|
|
- name: Configure Git safe directory
|
|
shell: bash
|
|
run: |
|
|
git config --global --add safe.directory $(pwd)
|
|
|
|
- uses: actions/setup-python@v5
|
|
with:
|
|
python-version: ${{ matrix.python-version }}
|
|
|
|
- uses: Jimver/cuda-toolkit@master
|
|
with:
|
|
cuda: ${{ matrix.cuda-version }}
|
|
sub-packages: '["nvcc", "toolkit"]'
|
|
method: "network"
|
|
use-github-cache: false
|
|
use-local-cache: false
|
|
|
|
- name: Install build dependencies
|
|
shell: bash
|
|
run: |
|
|
sudo apt install -y ninja-build clang
|
|
pip install -U pip setuptools==75.8.0 wheel setuptools packaging psutil
|
|
|
|
- name: Set environment variables
|
|
shell: bash
|
|
run: |
|
|
export PATH=/usr/local/nvidia/bin:/usr/local/nvidia/lib64:$PATH
|
|
export LD_LIBRARY_PATH=/usr/local/nvidia/lib64:/usr/local/cuda/lib64:$LD_LIBRARY_PATH
|
|
|
|
- name: Build wheels
|
|
timeout-minutes: 800
|
|
shell: bash
|
|
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" >> $GITHUB_ENV
|
|
|
|
- name: Install Test
|
|
shell: bash
|
|
run: |
|
|
pip install flash-attention/dist/${{ env.wheel_name }}
|
|
python -c "import flash_attn; print(flash_attn.__version__)"
|
|
|
|
- name: Upload Release Asset
|
|
shell: bash
|
|
env:
|
|
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
|
|
run: |
|
|
tag_name=${{ github.ref_name }}
|
|
wheel_path="flash-attention/dist/${{ env.wheel_name }}"
|
|
|
|
# Check if the file exists
|
|
if [ ! -f "$wheel_path" ]; then
|
|
echo "Error: Wheel file not found at $wheel_path"
|
|
exit 1
|
|
fi
|
|
|
|
# Upload the release asset using GitHub CLI
|
|
gh release upload "$tag_name" "$wheel_path" --clobber
|
|
|
|
echo "Successfully uploaded $wheel_name to release $tag_name"
|
|
|
|
- name: Clean up
|
|
shell: bash
|
|
if: always()
|
|
run: |
|
|
sudo rm -rf /opt/hostedtoolcache/Python/${{ matrix.python-version }}*
|
|
|
|
# build_windows_wheels:
|
|
# name: Build wheels and Upload (Windows x86_64, GitHub hosted runner)
|
|
# runs-on: windows-latest
|
|
# timeout-minutes: 1000
|
|
# env:
|
|
# MAX_JOBS: 2
|
|
# NVCC_THREADS: 2
|
|
# strategy:
|
|
# fail-fast: false
|
|
# matrix:
|
|
# flash-attn-version: ["2.5.9", "2.6.3", "2.7.4"]
|
|
# python-version: ["3.10", "3.11", "3.12"]
|
|
# torch-version: ["2.4.1", "2.5.1", "2.6.0", "2.7.1"]
|
|
# # https://developer.nvidia.com/cuda-toolkit-archive
|
|
# # CUDA 11.8 cannot build flash-attn.
|
|
# # https://github.com/Dao-AILab/flash-attention/issues/595
|
|
# cuda-version: ["12.4.1", "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: Enable Git long paths
|
|
# shell: pwsh
|
|
# run: git config --system core.longpaths true
|
|
|
|
# - uses: actions/setup-python@v5
|
|
# with:
|
|
# python-version: ${{ matrix.python-version }}
|
|
|
|
# - uses: Jimver/cuda-toolkit@v0.2.24
|
|
# with:
|
|
# cuda: ${{ matrix.cuda-version }}
|
|
# method: "network"
|
|
# use-github-cache: false
|
|
# use-local-cache: false
|
|
|
|
# - name: Install VS2022 BuildTools
|
|
# shell: pwsh
|
|
# run: |
|
|
# choco install -y visualstudio2022buildtools `
|
|
# --version=117.14.1 `
|
|
# --params "--add Microsoft.VisualStudio.Component.VC.Tools.x86.x64"
|
|
|
|
# - name: Install build dependencies
|
|
# shell: pwsh
|
|
# run: |
|
|
# pip install -U pip setuptools==75.8.0 wheel setuptools packaging psutil ninja
|
|
|
|
# - name: Build wheels
|
|
# shell: pwsh
|
|
# run: |
|
|
# .\build_windows.ps1 -FlashAttnVersion "${{ matrix.flash-attn-version }}" -PythonVersion "${{ matrix.python-version }}" -TorchVersion "${{ matrix.torch-version }}" -CudaVersion "${{ matrix.cuda-version }}"
|
|
# $wheelName = Get-ChildItem -Path "flash-attention\dist\*.whl" | Select-Object -First 1 | ForEach-Object { $_.Name }
|
|
# echo "wheel_name=$wheelName" >> $env:GITHUB_ENV
|
|
|
|
# - name: Install Test
|
|
# shell: pwsh
|
|
# run: |
|
|
# pip install flash-attention/dist/$env:wheel_name
|
|
# python -c "import flash_attn; print(flash_attn.__version__)"
|
|
|
|
# - name: Upload Release Asset
|
|
# shell: pwsh
|
|
# env:
|
|
# GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
|
|
# run: |
|
|
# $tag_name = "${env:GITHUB_REF}".Replace("refs/tags/", "")
|
|
# $wheel_path = "flash-attention/dist/$env:wheel_name"
|
|
|
|
# # Check if the file exists
|
|
# if (-not (Test-Path $wheel_path)) {
|
|
# $tag_name = "${env:GITHUB_REF}".Replace("refs/tags/", "")
|
|
# $wheel_path = "flash-attention/dist/$env:wheel_name"
|
|
|
|
# # Check if the file exists
|
|
# if (-not (Test-Path $wheel_path)) {
|
|
# Write-Host "Error: Wheel file not found at $wheel_path"
|
|
# exit 1
|
|
# }
|
|
|
|
# # Upload the release asset using GitHub CLI
|
|
# gh release upload "$tag_name" "$wheel_path" --clobber
|
|
|
|
# Write-Host "Successfully uploaded $env:wheel_name to release $tag_name"
|