--- language: - en tags: - flash_attention_2 - pytorch - flash-attn - flash-attn @ <> - text-generation-inference license: bsd-3-clause library_name: transformers --- ![flashattention_logo](https://cdn-uploads.huggingface.co/production/uploads/65bb837dbfb878f46c77de4c/yQ8ugTupP2yLUV9sZ4HXt.png) `flash-attn @ <>` # **Flash Attention 2 Pre-built Wheels** > [!IMPORTANT] > This repository contains a copy of all prebuilt wheels for **flash_attn-2.8.3** from [Dao-AILab's Flash Attention](https://github.com/Dao-AILab/flash-attention), organized for better categorization based on PyTorch versions and Python (cp) release versions. This repository provides pre-built wheels for `flash-attn` version **2.8.3** for various PyTorch versions, Python versions, and architectures (compiled with CUDA 12). You can install these directly using `pip install ` or add the provided strings directly to your `requirements.txt`. > [!IMPORTANT] > The detailed categories and structured view of the `strangertoolshf/flash_attention_2_wheelhouse` folders and files on the hf-tree shareable link are available here: [huggingface-tree](https://strangertoolshf-huggingface-tree.hf.space/#models/strangertoolshf/flash_attention_2_wheelhouse/main)
Open this to view the entire directory. ```py strangertoolshf/flash_attention_2_wheelhouse (main) +-- wheelhouse-flash_attn-2.8.3 | +-- linux_aarch64 | | +-- torch2.9 | | +-- cu12 | | +-- abiTRUE | | +-- cp312 | | +-- flash_attn-2.8.3+cu12torch2.9cxx11abiTRUE-cp312-cp312-linux_aarch64.whl (241.8 MB) | +-- linux_x86_64 | +-- torch2.4 | | +-- cu12 | | +-- abiFALSE | | | +-- cp310 | | | | +-- flash_attn-2.8.3+cu12torch2.4cxx11abiFALSE-cp310-cp310-linux_x86_64.whl (244.1 MB) | | | +-- cp311 | | | | +-- flash_attn-2.8.3+cu12torch2.4cxx11abiFALSE-cp311-cp311-linux_x86_64.whl (244.1 MB) | | | +-- cp312 | | | | +-- flash_attn-2.8.3+cu12torch2.4cxx11abiFALSE-cp312-cp312-linux_x86_64.whl (244.1 MB) | | | +-- cp39 | | | +-- flash_attn-2.8.3+cu12torch2.4cxx11abiFALSE-cp39-cp39-linux_x86_64.whl (244.1 MB) | | +-- abiTRUE | | +-- cp310 | | | +-- flash_attn-2.8.3+cu12torch2.4cxx11abiTRUE-cp310-cp310-linux_x86_64.whl (244.1 MB) | | +-- cp311 | | | +-- flash_attn-2.8.3+cu12torch2.4cxx11abiTRUE-cp311-cp311-linux_x86_64.whl (244.1 MB) | | +-- cp312 | | | +-- flash_attn-2.8.3+cu12torch2.4cxx11abiTRUE-cp312-cp312-linux_x86_64.whl (244.1 MB) | | +-- cp39 | | +-- flash_attn-2.8.3+cu12torch2.4cxx11abiTRUE-cp39-cp39-linux_x86_64.whl (244.1 MB) | +-- torch2.5 | | +-- cu12 | | +-- abiFALSE | | | +-- cp310 | | | | +-- flash_attn-2.8.3+cu12torch2.5cxx11abiFALSE-cp310-cp310-linux_x86_64.whl (244.1 MB) | | | +-- cp311 | | | | +-- flash_attn-2.8.3+cu12torch2.5cxx11abiFALSE-cp311-cp311-linux_x86_64.whl (244.1 MB) | | | +-- cp312 | | | | +-- flash_attn-2.8.3+cu12torch2.5cxx11abiFALSE-cp312-cp312-linux_x86_64.whl (244.1 MB) | | | +-- cp313 | | | | +-- flash_attn-2.8.3+cu12torch2.5cxx11abiFALSE-cp313-cp313-linux_x86_64.whl (244.1 MB) | | | +-- cp39 | | | +-- flash_attn-2.8.3+cu12torch2.5cxx11abiFALSE-cp39-cp39-linux_x86_64.whl (244.1 MB) | | +-- abiTRUE | | +-- cp310 | | | +-- flash_attn-2.8.3+cu12torch2.5cxx11abiTRUE-cp310-cp310-linux_x86_64.whl (244.1 MB) | | +-- cp311 | | | +-- flash_attn-2.8.3+cu12torch2.5cxx11abiTRUE-cp311-cp311-linux_x86_64.whl (244.2 MB) | | +-- cp312 | | | +-- flash_attn-2.8.3+cu12torch2.5cxx11abiTRUE-cp312-cp312-linux_x86_64.whl (244.2 MB) | | +-- cp313 | | | +-- flash_attn-2.8.3+cu12torch2.5cxx11abiTRUE-cp313-cp313-linux_x86_64.whl (244.2 MB) | | +-- cp39 | | +-- flash_attn-2.8.3+cu12torch2.5cxx11abiTRUE-cp39-cp39-linux_x86_64.whl (244.1 MB) | +-- torch2.6 | | +-- cu12 | | +-- abiFALSE | | | +-- cp310 | | | | +-- flash_attn-2.8.3+cu12torch2.6cxx11abiFALSE-cp310-cp310-linux_x86_64.whl (244.1 MB) | | | +-- cp311 | | | | +-- flash_attn-2.8.3+cu12torch2.6cxx11abiFALSE-cp311-cp311-linux_x86_64.whl (244.2 MB) | | | +-- cp312 | | | | +-- flash_attn-2.8.3+cu12torch2.6cxx11abiFALSE-cp312-cp312-linux_x86_64.whl (244.2 MB) | | | +-- cp313 | | | | +-- flash_attn-2.8.3+cu12torch2.6cxx11abiFALSE-cp313-cp313-linux_x86_64.whl (244.2 MB) | | | +-- cp39 | | | +-- flash_attn-2.8.3+cu12torch2.6cxx11abiFALSE-cp39-cp39-linux_x86_64.whl (244.1 MB) | | +-- abiTRUE | | +-- cp310 | | | +-- flash_attn-2.8.3+cu12torch2.6cxx11abiTRUE-cp310-cp310-linux_x86_64.whl (244.1 MB) | | +-- cp311 | | | +-- flash_attn-2.8.3+cu12torch2.6cxx11abiTRUE-cp311-cp311-linux_x86_64.whl (244.2 MB) | | +-- cp312 | | | +-- flash_attn-2.8.3+cu12torch2.6cxx11abiTRUE-cp312-cp312-linux_x86_64.whl (244.2 MB) | | +-- cp313 | | | +-- flash_attn-2.8.3+cu12torch2.6cxx11abiTRUE-cp313-cp313-linux_x86_64.whl (244.2 MB) | | +-- cp39 | | +-- flash_attn-2.8.3+cu12torch2.6cxx11abiTRUE-cp39-cp39-linux_x86_64.whl (244.1 MB) | +-- torch2.7 | | +-- cu12 | | +-- abiFALSE | | | +-- cp310 | | | | +-- flash_attn-2.8.3+cu12torch2.7cxx11abiFALSE-cp310-cp310-linux_x86_64.whl (244.1 MB) | | | +-- cp311 | | | | +-- flash_attn-2.8.3+cu12torch2.7cxx11abiFALSE-cp311-cp311-linux_x86_64.whl (244.2 MB) | | | +-- cp312 | | | | +-- flash_attn-2.8.3+cu12torch2.7cxx11abiFALSE-cp312-cp312-linux_x86_64.whl (244.2 MB) | | | +-- cp313 | | | | +-- flash_attn-2.8.3+cu12torch2.7cxx11abiFALSE-cp313-cp313-linux_x86_64.whl (244.2 MB) | | | +-- cp39 | | | +-- flash_attn-2.8.3+cu12torch2.7cxx11abiFALSE-cp39-cp39-linux_x86_64.whl (244.1 MB) | | +-- abiTRUE | | +-- cp310 | | | +-- flash_attn-2.8.3+cu12torch2.7cxx11abiTRUE-cp310-cp310-linux_x86_64.whl (244.1 MB) | | +-- cp311 | | | +-- flash_attn-2.8.3+cu12torch2.7cxx11abiTRUE-cp311-cp311-linux_x86_64.whl (244.2 MB) | | +-- cp312 | | | +-- flash_attn-2.8.3+cu12torch2.7cxx11abiTRUE-cp312-cp312-linux_x86_64.whl (244.2 MB) | | +-- cp313 | | | +-- flash_attn-2.8.3+cu12torch2.7cxx11abiTRUE-cp313-cp313-linux_x86_64.whl (244.2 MB) | | +-- cp39 | | +-- flash_attn-2.8.3+cu12torch2.7cxx11abiTRUE-cp39-cp39-linux_x86_64.whl (244.1 MB) | +-- torch2.8 | | +-- cu12 | | +-- abiFALSE | | | +-- cp310 | | | | +-- flash_attn-2.8.3+cu12torch2.8cxx11abiFALSE-cp310-cp310-linux_x86_64.whl (244.2 MB) | | | +-- cp311 | | | | +-- flash_attn-2.8.3+cu12torch2.8cxx11abiFALSE-cp311-cp311-linux_x86_64.whl (244.2 MB) | | | +-- cp312 | | | | +-- flash_attn-2.8.3+cu12torch2.8cxx11abiFALSE-cp312-cp312-linux_x86_64.whl (244.2 MB) | | | +-- cp313 | | | | +-- flash_attn-2.8.3+cu12torch2.8cxx11abiFALSE-cp313-cp313-linux_x86_64.whl (244.2 MB) | | | +-- cp39 | | | +-- flash_attn-2.8.3+cu12torch2.8cxx11abiFALSE-cp39-cp39-linux_x86_64.whl (244.2 MB) | | +-- abiTRUE | | +-- cp310 | | | +-- flash_attn-2.8.3+cu12torch2.8cxx11abiTRUE-cp310-cp310-linux_x86_64.whl (244.2 MB) | | +-- cp311 | | | +-- flash_attn-2.8.3+cu12torch2.8cxx11abiTRUE-cp311-cp311-linux_x86_64.whl (244.2 MB) | | +-- cp312 | | | +-- flash_attn-2.8.3+cu12torch2.8cxx11abiTRUE-cp312-cp312-linux_x86_64.whl (244.2 MB) | | +-- cp313 | | | +-- flash_attn-2.8.3+cu12torch2.8cxx11abiTRUE-cp313-cp313-linux_x86_64.whl (244.2 MB) | | +-- cp39 | | +-- flash_attn-2.8.3+cu12torch2.8cxx11abiTRUE-cp39-cp39-linux_x86_64.whl (244.2 MB) | +-- torch2.9 | +-- cu12 | +-- cp312 | +-- flash_attn-2.8.3+cu12torch2.9cxx11abiTRUE-cp312-cp312-linux_x86_64.whl (242.0 MB) +-- .gitattributes (10.1 KB) +-- flash-attention-2.8.3.zip (6.2 MB) +-- README.md (16.9 KB) ```
--- ## **Quick Start with StrangerTools-HF** ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-//torch/cu/abi//flash_attn-+cutorchcxx11abi---.whl ``` ### Breakdown of the Variables: * **``**: The version of Flash Attention (e.g., `2.8.3`). * **``**: The OS and system architecture (e.g., `linux_x86_64`, `linux_aarch64`). * **``**: The PyTorch version without the minor patch (e.g., `2.4`, `2.5`, `2.8`). * **``**: The CUDA major version (e.g., `12`). * **``**: The CXX11 ABI boolean flag (e.g., `TRUE`, `FALSE`). * **``**: The specific CPython version tag (e.g., `cp39`, `cp310`, `cp312`). > [!IMPORTANT] Find all the provided pre-built wheels below.👇 --- ## Linux x86_64 ### **Torch 2.9** **ABI: `TRUE` (Implied)** #### Python 3.12 (`cp312`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.9/cu12/cp312/flash_attn-2.8.3+cu12torch2.9cxx11abiTRUE-cp312-cp312-linux_x86_64.whl ``` > End of wheels for Torch 2.9. --- ### **Torch 2.8** #### ABI: `FALSE` ##### Python 3.9 (`cp39`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.8/cu12/abiFALSE/cp39/flash_attn-2.8.3+cu12torch2.8cxx11abiFALSE-cp39-cp39-linux_x86_64.whl ``` ##### Python 3.10 (`cp310`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.8/cu12/abiFALSE/cp310/flash_attn-2.8.3+cu12torch2.8cxx11abiFALSE-cp310-cp310-linux_x86_64.whl ``` ##### Python 3.11 (`cp311`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.8/cu12/abiFALSE/cp311/flash_attn-2.8.3+cu12torch2.8cxx11abiFALSE-cp311-cp311-linux_x86_64.whl ``` ##### Python 3.12 (`cp312`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.8/cu12/abiFALSE/cp312/flash_attn-2.8.3+cu12torch2.8cxx11abiFALSE-cp312-cp312-linux_x86_64.whl ``` ##### Python 3.13 (`cp313`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.8/cu12/abiFALSE/cp313/flash_attn-2.8.3+cu12torch2.8cxx11abiFALSE-cp313-cp313-linux_x86_64.whl ``` #### ABI: `TRUE` ##### Python 3.9 (`cp39`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.8/cu12/abiTRUE/cp39/flash_attn-2.8.3+cu12torch2.8cxx11abiTRUE-cp39-cp39-linux_x86_64.whl ``` ##### Python 3.10 (`cp310`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.8/cu12/abiTRUE/cp310/flash_attn-2.8.3+cu12torch2.8cxx11abiTRUE-cp310-cp310-linux_x86_64.whl ``` ##### Python 3.11 (`cp311`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.8/cu12/abiTRUE/cp311/flash_attn-2.8.3+cu12torch2.8cxx11abiTRUE-cp311-cp311-linux_x86_64.whl ``` ##### Python 3.12 (`cp312`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.8/cu12/abiTRUE/cp312/flash_attn-2.8.3+cu12torch2.8cxx11abiTRUE-cp312-cp312-linux_x86_64.whl ``` ##### Python 3.13 (`cp313`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.8/cu12/abiTRUE/cp313/flash_attn-2.8.3+cu12torch2.8cxx11abiTRUE-cp313-cp313-linux_x86_64.whl ``` --- ### **Torch 2.7** #### ABI: `FALSE` ##### Python 3.9 (`cp39`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.7/cu12/abiFALSE/cp39/flash_attn-2.8.3+cu12torch2.7cxx11abiFALSE-cp39-cp39-linux_x86_64.whl ``` ##### Python 3.10 (`cp310`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.7/cu12/abiFALSE/cp310/flash_attn-2.8.3+cu12torch2.7cxx11abiFALSE-cp310-cp310-linux_x86_64.whl ``` ##### Python 3.11 (`cp311`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.7/cu12/abiFALSE/cp311/flash_attn-2.8.3+cu12torch2.7cxx11abiFALSE-cp311-cp311-linux_x86_64.whl ``` ##### Python 3.12 (`cp312`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.7/cu12/abiFALSE/cp312/flash_attn-2.8.3+cu12torch2.7cxx11abiFALSE-cp312-cp312-linux_x86_64.whl ``` ##### Python 3.13 (`cp313`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.7/cu12/abiFALSE/cp313/flash_attn-2.8.3+cu12torch2.7cxx11abiFALSE-cp313-cp313-linux_x86_64.whl ``` #### ABI: `TRUE` ##### Python 3.9 (`cp39`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.7/cu12/abiTRUE/cp39/flash_attn-2.8.3+cu12torch2.7cxx11abiTRUE-cp39-cp39-linux_x86_64.whl ``` ##### Python 3.10 (`cp310`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.7/cu12/abiTRUE/cp310/flash_attn-2.8.3+cu12torch2.7cxx11abiTRUE-cp310-cp310-linux_x86_64.whl ``` ##### Python 3.11 (`cp311`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.7/cu12/abiTRUE/cp311/flash_attn-2.8.3+cu12torch2.7cxx11abiTRUE-cp311-cp311-linux_x86_64.whl ``` ##### Python 3.12 (`cp312`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.7/cu12/abiTRUE/cp312/flash_attn-2.8.3+cu12torch2.7cxx11abiTRUE-cp312-cp312-linux_x86_64.whl ``` ##### Python 3.13 (`cp313`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.7/cu12/abiTRUE/cp313/flash_attn-2.8.3+cu12torch2.7cxx11abiTRUE-cp313-cp313-linux_x86_64.whl ``` --- ### **Torch 2.6** #### ABI: `FALSE` ##### Python 3.9 (`cp39`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.6/cu12/abiFALSE/cp39/flash_attn-2.8.3+cu12torch2.6cxx11abiFALSE-cp39-cp39-linux_x86_64.whl ``` ##### Python 3.10 (`cp310`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.6/cu12/abiFALSE/cp310/flash_attn-2.8.3+cu12torch2.6cxx11abiFALSE-cp310-cp310-linux_x86_64.whl ``` ##### Python 3.11 (`cp311`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.6/cu12/abiFALSE/cp311/flash_attn-2.8.3+cu12torch2.6cxx11abiFALSE-cp311-cp311-linux_x86_64.whl ``` ##### Python 3.12 (`cp312`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.6/cu12/abiFALSE/cp312/flash_attn-2.8.3+cu12torch2.6cxx11abiFALSE-cp312-cp312-linux_x86_64.whl ``` ##### Python 3.13 (`cp313`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.6/cu12/abiFALSE/cp313/flash_attn-2.8.3+cu12torch2.6cxx11abiFALSE-cp313-cp313-linux_x86_64.whl ``` #### ABI: `TRUE` ##### Python 3.9 (`cp39`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.6/cu12/abiTRUE/cp39/flash_attn-2.8.3+cu12torch2.6cxx11abiTRUE-cp39-cp39-linux_x86_64.whl ``` ##### Python 3.10 (`cp310`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.6/cu12/abiTRUE/cp310/flash_attn-2.8.3+cu12torch2.6cxx11abiTRUE-cp310-cp310-linux_x86_64.whl ``` ##### Python 3.11 (`cp311`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.6/cu12/abiTRUE/cp311/flash_attn-2.8.3+cu12torch2.6cxx11abiTRUE-cp311-cp311-linux_x86_64.whl ``` ##### Python 3.12 (`cp312`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.6/cu12/abiTRUE/cp312/flash_attn-2.8.3+cu12torch2.6cxx11abiTRUE-cp312-cp312-linux_x86_64.whl ``` ##### Python 3.13 (`cp313`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.6/cu12/abiTRUE/cp313/flash_attn-2.8.3+cu12torch2.6cxx11abiTRUE-cp313-cp313-linux_x86_64.whl ``` --- ### **Torch 2.5** #### ABI: `FALSE` ##### Python 3.9 (`cp39`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.5/cu12/abiFALSE/cp39/flash_attn-2.8.3+cu12torch2.5cxx11abiFALSE-cp39-cp39-linux_x86_64.whl ``` ##### Python 3.10 (`cp310`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.5/cu12/abiFALSE/cp310/flash_attn-2.8.3+cu12torch2.5cxx11abiFALSE-cp310-cp310-linux_x86_64.whl ``` ##### Python 3.11 (`cp311`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.5/cu12/abiFALSE/cp311/flash_attn-2.8.3+cu12torch2.5cxx11abiFALSE-cp311-cp311-linux_x86_64.whl ``` ##### Python 3.12 (`cp312`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.5/cu12/abiFALSE/cp312/flash_attn-2.8.3+cu12torch2.5cxx11abiFALSE-cp312-cp312-linux_x86_64.whl ``` ##### Python 3.13 (`cp313`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.5/cu12/abiFALSE/cp313/flash_attn-2.8.3+cu12torch2.5cxx11abiFALSE-cp313-cp313-linux_x86_64.whl ``` #### ABI: `TRUE` ##### Python 3.9 (`cp39`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.5/cu12/abiTRUE/cp39/flash_attn-2.8.3+cu12torch2.5cxx11abiTRUE-cp39-cp39-linux_x86_64.whl ``` ##### Python 3.10 (`cp310`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.5/cu12/abiTRUE/cp310/flash_attn-2.8.3+cu12torch2.5cxx11abiTRUE-cp310-cp310-linux_x86_64.whl ``` ##### Python 3.11 (`cp311`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.5/cu12/abiTRUE/cp311/flash_attn-2.8.3+cu12torch2.5cxx11abiTRUE-cp311-cp311-linux_x86_64.whl ``` ##### Python 3.12 (`cp312`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.5/cu12/abiTRUE/cp312/flash_attn-2.8.3+cu12torch2.5cxx11abiTRUE-cp312-cp312-linux_x86_64.whl ``` ##### Python 3.13 (`cp313`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.5/cu12/abiTRUE/cp313/flash_attn-2.8.3+cu12torch2.5cxx11abiTRUE-cp313-cp313-linux_x86_64.whl ``` --- ### **Torch 2.4** #### ABI: `FALSE` ##### Python 3.9 (`cp39`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.4/cu12/abiFALSE/cp39/flash_attn-2.8.3+cu12torch2.4cxx11abiFALSE-cp39-cp39-linux_x86_64.whl ``` ##### Python 3.10 (`cp310`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.4/cu12/abiFALSE/cp310/flash_attn-2.8.3+cu12torch2.4cxx11abiFALSE-cp310-cp310-linux_x86_64.whl ``` ##### Python 3.11 (`cp311`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.4/cu12/abiFALSE/cp311/flash_attn-2.8.3+cu12torch2.4cxx11abiFALSE-cp311-cp311-linux_x86_64.whl ``` ##### Python 3.12 (`cp312`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.4/cu12/abiFALSE/cp312/flash_attn-2.8.3+cu12torch2.4cxx11abiFALSE-cp312-cp312-linux_x86_64.whl ``` #### ABI: `TRUE` ##### Python 3.9 (`cp39`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.4/cu12/abiTRUE/cp39/flash_attn-2.8.3+cu12torch2.4cxx11abiTRUE-cp39-cp39-linux_x86_64.whl ``` ##### Python 3.10 (`cp310`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.4/cu12/abiTRUE/cp310/flash_attn-2.8.3+cu12torch2.4cxx11abiTRUE-cp310-cp310-linux_x86_64.whl ``` ##### Python 3.11 (`cp311`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.4/cu12/abiTRUE/cp311/flash_attn-2.8.3+cu12torch2.4cxx11abiTRUE-cp311-cp311-linux_x86_64.whl ``` ##### Python 3.12 (`cp312`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_x86_64/torch2.4/cu12/abiTRUE/cp312/flash_attn-2.8.3+cu12torch2.4cxx11abiTRUE-cp312-cp312-linux_x86_64.whl ``` --- ## Linux aarch64 (ARM) ### **Torch 2.9 (Nightly/Pre-release)** #### ABI: `TRUE` ##### Python 3.12 (`cp312`) ``` flash-attn @ https://huggingface.co/strangertoolshf/flash_attention_2_wheelhouse/resolve/main/wheelhouse-flash_attn-2.8.3/linux_aarch64/torch2.9/cu12/abiTRUE/cp312/flash_attn-2.8.3+cu12torch2.9cxx11abiTRUE-cp312-cp312-linux_aarch64.whl ``` --- ## Acknowledgements and Note: - Dao-AILab [flash-attention]: [Faster Attention with Better Parallelism and Work Partitioning](https://arxiv.org/pdf/2205.14135). - Note: This repository follows the same license, release notices, and other terms and conditions as the Dao-AILab [flash-attention](https://github.com/Dao-AILab/flash-attention) repository.