flash-attn quick installer
Project description
⚡ qifa – Quick Installer for Flash Attention
One command to detect, select, and install the correct flash-attn wheel for your environment.
✨ Features
- Auto-Detection – Identifies Python version, PyTorch version, CUDA version, and C++11 ABI
- Smart Matching – Finds the correct prebuilt wheel from official GitHub Releases
- Zero Config – Just run
qifa installand you're done - Dry Run Mode – Preview what will be installed before committing
Note: Currently optimized for Linux x86_64 official wheels. Windows/ROCm users will receive guidance but may need manual installation.
📦 Installation
pip install qifa
🚀 Quick Start
# Check your environment and see which wheel will be installed
qifa plan
# Install the matching flash-attn wheel
qifa install
📖 Commands
| Command | Description |
|---|---|
qifa plan |
Show detected environment and the wheel that will be installed |
qifa install |
Download and install the matching wheel |
qifa uninstall |
Remove existing flash-attn installation |
qifa doctor |
Run compatibility checks and get guidance |
Options
# Install a specific version
qifa install --version 2.5.8
# Force a specific C++11 ABI setting
qifa install --abi FALSE
# Preview without installing
qifa install --dry-run
🔍 Example Output
$ qifa plan
{
"python_tag": "cp310",
"torch_mm": "2.1",
"torch_cuda": "cu121",
"chosen_cu_tag": "cu121",
"abi_FALSE": true,
"platform": "linux_x86_64",
"version": "2.5.8",
"found_asset": "flash_attn-2.5.8+cu121torch2.1cxx11abiFALSE-cp310-cp310-linux_x86_64.whl",
"download_url": "https://github.com/Dao-AILab/flash-attention/releases/download/..."
}
🛠️ Troubleshooting
| Issue | Solution |
|---|---|
| No matching wheel found | Try --version to specify a different version, or --abi TRUE |
| PyTorch not detected | Install CUDA-enabled PyTorch first: pip install torch --index-url https://download.pytorch.org/whl/cu121 |
| Platform not supported | Official wheels are for Linux x86_64; other platforms require building from source |
Run qifa doctor for detailed guidance on your specific environment.
📄 License
BSD-3-Clause © Voidful
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file qifa-0.1.1.tar.gz.
File metadata
- Download URL: qifa-0.1.1.tar.gz
- Upload date:
- Size: 5.6 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.12.9
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
b813aa0d080e9f4bc4835de92edf571103aaa0316bb9574b1364a85427e9df55
|
|
| MD5 |
e6d0bbaac0166cfaaae5f08801c638e7
|
|
| BLAKE2b-256 |
c67aacf9dc4ff60b5709c192fa447b108a761bd370c478406070282da8a86cd3
|
File details
Details for the file qifa-0.1.1-py3-none-any.whl.
File metadata
- Download URL: qifa-0.1.1-py3-none-any.whl
- Upload date:
- Size: 6.0 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.12.9
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
17c0f843e27e6e024cd41f2170052e18614ab0f343530efe03fa979da9014f5e
|
|
| MD5 |
261448bc60556ffd0258bf9933e3ce54
|
|
| BLAKE2b-256 |
ba5217e11d665196e3f8c121752b90564d4a5e7ce726be883f617086656df0d8
|