yindo/flash-attention-prebuild-wheels-rocm
1
0
Fork 0
mirror of https://github.com/BillyOutlast/flash-attention-prebuild-wheels-rocm.git synced 2026-07-01 01:37:53 -04:00
Code Issues Packages Projects Releases 2 Wiki Activity

update workflow

Browse Source
This commit is contained in:
Junya Morioka
2025-08-10 16:47:51 +09:00
parent 03dfdc4389
commit 648fc9c600
6 changed files with 169 additions and 324 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.github/workflows/build.yml
+42 -275
View File
@@ -17,29 +17,14 @@ jobs:
run: |
gh release create "${{ github.ref_name }}" \
--title "${{ github.ref_name }}" \
--notes "## Linux x86_64
| Flash-Attention | Python | PyTorch | CUDA |
| --- | --- | --- | --- |
| 2.7.4 | 3.10, 3.11, 3.12 | 2.5.1, 2.6.0, 2.7.1, 2.8.0 | 12.4.1, 12.8.1, 12.9.1 |
## Windows x86_64
| Flash-Attention | Python | PyTorch | CUDA |
| --- | --- | --- | --- |
| 2.6.3, 2.7.4, 2.8.2 | 3.10, 3.11, 3.12 | 2.5.1, 2.6.0, 2.7.1, 2.8.0 | 12.4.1, 12.8.1 |
"
--notes "TBD"
# #########################################################
# Build wheels with GitHub hosted runner
# #########################################################
build_wheels:
name: Build wheels and Upload (Linux x86_64, GitHub hosted runner)
build_wheels_linux:
name: Build Linux
needs: create_releases
runs-on: ubuntu-22.04
env:
DEBIAN_FRONTEND: noninteractive
TERM: xterm-256color
strategy:
fail-fast: false
matrix:
@@ -76,93 +61,19 @@ jobs:
cuda-version: "12.9.1"
- torch-version: "2.7.1"
cuda-version: "12.9.1"
steps:
- uses: actions/checkout@v4
# flash-attn 2.7.4 does not build in GitHub Hosted Runner
- flash-attn-version: "2.7.4"
uses: ./.github/workflows/build_linux.yml
with:
flash-attn-version: ${{ matrix.flash-attn-version }}
python-version: ${{ matrix.python-version }}
torch-version: ${{ matrix.torch-version }}
cuda-version: ${{ matrix.cuda-version }}
secrets: inherit
- 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
env:
MAX_JOBS: 2
NVCC_THREADS: 2
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" >> $GITHUB_ENV
- name: Install Test
run: |
pip install --no-cache-dir flash-attention/dist/${{ 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
# #########################################################
# Build wheels with self-hosted runner
# #########################################################
build_wheels_self_hosted:
name: Build wheels and Upload (Linux x86_64, self-hosted runner)
build_wheels_linux_self_hosted:
name: Build Linux (self-hosted)
needs: create_releases
runs-on: self-hosted
container:
image: ubuntu:22.04
defaults:
run:
shell: bash
env:
DEBIAN_FRONTEND: noninteractive
TERM: xterm-256color
timeout-minutes: 2000
strategy:
fail-fast: false
matrix:
@@ -192,9 +103,6 @@ jobs:
# torch 2.7.0 does not support CUDA 12.4
- torch-version: "2.7.0"
cuda-version: "12.4.1"
# flash-attn 2.7.4 does not build with CUDA 12.8
- flash-attn-version: "2.7.4"
cuda-version: "12.8.1"
# torch < 2.8 does not support CUDA 12.9
- torch-version: "2.5.1"
cuda-version: "12.9.1"
@@ -202,119 +110,17 @@ jobs:
cuda-version: "12.9.1"
- torch-version: "2.7.1"
cuda-version: "12.9.1"
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: ./.github/workflows/build_linux_self_host.yml
with:
flash-attn-version: ${{ matrix.flash-attn-version }}
python-version: ${{ matrix.python-version }}
torch-version: ${{ matrix.torch-version }}
cuda-version: ${{ matrix.cuda-version }}
secrets: inherit
- name: Install gh
shell: bash
run: |
sudo mkdir -p -m 755 /etc/apt/keyrings
out=$(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: 1200
shell: bash
env:
MAX_JOBS: 2
NVCC_THREADS: 2
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" >> $GITHUB_ENV
- name: Install Test
shell: bash
run: |
pip uninstall -y flash-attn > /dev/null 2>&1
pip install --no-cache-dir flash-attention/dist/${{ 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
- name: Clean up
shell: bash
if: always()
run: |
rm -rf /opt/hostedtoolcache/Python
build_windows_wheels:
name: Build wheels and Upload (Windows x86_64, GitHub hosted runner)
runs-on: windows-2022
timeout-minutes: 1000
build_wheels_windows:
name: Build Windows
needs: create_releases
env:
MAX_JOBS: 2
NVCC_THREADS: 2
strategy:
fail-fast: false
matrix:
@@ -344,9 +150,6 @@ jobs:
# torch 2.7.0 does not support CUDA 12.4
- torch-version: "2.7.0"
cuda-version: "12.4.1"
# flash-attn 2.7.4 does not build with CUDA 12.8
- flash-attn-version: "2.7.4"
cuda-version: "12.8.1"
# torch < 2.8 does not support CUDA 12.9
- torch-version: "2.5.1"
cuda-version: "12.9.1"
@@ -354,67 +157,31 @@ jobs:
cuda-version: "12.9.1"
- torch-version: "2.7.1"
cuda-version: "12.9.1"
uses: ./.github/workflows/build_windows.yml
with:
flash-attn-version: ${{ matrix.flash-attn-version }}
python-version: ${{ matrix.python-version }}
torch-version: ${{ matrix.torch-version }}
cuda-version: ${{ matrix.cuda-version }}
secrets: inherit
update_release_notes:
name: Update Release Notes
needs:
[build_wheels_linux, build_wheels_linux_self_hosted, build_wheels_windows]
if: always()
runs-on: ubuntu-latest
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 }}
python-version: 3.12
- uses: Jimver/cuda-toolkit@v0.2.24
with:
cuda: ${{ matrix.cuda-version }}
method: "network"
use-github-cache: false
use-local-cache: false
- name: Install VS2022 BuildTools
shell: pwsh
run: |
choco install -y visualstudio2022buildtools `
--version=117.14.1 `
--params "--add Microsoft.VisualStudio.Component.VC.Tools.x86.x64"
- name: Install build dependencies
shell: pwsh
run: |
python -m pip install -U setuptools==75.8.0 wheel packaging psutil ninja
- name: Build wheels
shell: pwsh
run: |
.\build_windows.ps1 -FlashAttnVersion "${{ 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" >> $env:GITHUB_ENV
- name: Install Test
shell: pwsh
run: |
pip install --no-cache-dir flash-attention/dist/$env:wheel_name
python -c "import flash_attn; print(flash_attn.__version__)"
- name: Upload Release Asset
shell: pwsh
- name: Generate Release Notes
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" "$wheel_path" --clobber
gh release view "${{ github.ref_name }}" --json assets > /tmp/assets.json
python create_release_note.py /tmp/assets.json > /tmp/release_notes.md
gh release edit "${{ github.ref_name }}" --notes-file /tmp/release_notes.md
.github/workflows/build_linux.yml
+1 -1
View File
@@ -2,7 +2,7 @@
# Build wheels with GitHub hosted runner
# #########################################################
name: Build wheels and upload to GitHub Releases (Linux x86_64)
name: "[Linux x86_64] Build wheels and upload to GitHub Releases"
on:
workflow_call:
.github/workflows/build_linux_self_host.yml
+3 -3
View File
@@ -2,7 +2,7 @@
# Build wheels with self-hosted runner
# #########################################################
name: Build wheels and upload to GitHub Releases (Linux x86_64, self-hosted runner)
name: "[Linux x86_64, self-hosted runner] Build wheels and upload to GitHub Releases"
on:
workflow_call:
@@ -105,8 +105,8 @@ jobs:
id: build_wheels
shell: bash
env:
MAX_JOBS: 2
NVCC_THREADS: 2
MAX_JOBS: 3
NVCC_THREADS: 3
run: |
chmod +x build_linux.sh
./build_linux.sh ${{ inputs.flash-attn-version }} ${{ inputs.python-version }} ${{ inputs.torch-version }} ${{ inputs.cuda-version }}
.github/workflows/build_test.yml
-44
View File
@@ -1,44 +0,0 @@
name: Build wheels and upload to GitHub Releases
on:
workflow_dispatch:
jobs:
# #########################################################
# Build wheels with GitHub hosted runner
# #########################################################
build_wheels_linux:
name: Build Linux
strategy:
fail-fast: false
matrix:
flash-attn-version: ["2.6.3"]
python-version: ["3.10"]
torch-version: ["2.5.1"]
# https://developer.nvidia.com/cuda-toolkit-archive
cuda-version: ["12.4.1"]
uses: ./.github/workflows/build_linux.yml
with:
flash-attn-version: ${{ matrix.flash-attn-version }}
python-version: ${{ matrix.python-version }}
torch-version: ${{ matrix.torch-version }}
cuda-version: ${{ matrix.cuda-version }}
secrets: inherit
build_wheels_windows:
name: Build Windows
strategy:
fail-fast: false
matrix:
flash-attn-version: ["2.6.3"]
python-version: ["3.10"]
torch-version: ["2.5.1"]
# https://developer.nvidia.com/cuda-toolkit-archive
cuda-version: ["12.4.1"]
uses: ./.github/workflows/build_windows.yml
with:
flash-attn-version: ${{ matrix.flash-attn-version }}
python-version: ${{ matrix.python-version }}
torch-version: ${{ matrix.torch-version }}
cuda-version: ${{ matrix.cuda-version }}
secrets: inherit
.github/workflows/build_windows.yml
+1 -1
View File
@@ -2,7 +2,7 @@
# Build wheels with GitHub hosted runner on Windows x86_64
# #########################################################
name: Build wheels and upload to GitHub Releases (Windows x86_64)
name: "[Windows x86_64] Build wheels and upload to GitHub Releases"
on:
workflow_call: