flash-attn 2.8.3 | CUDA 12 | PyTorch 2.11 | Python 3.10-3.13

Pre-built community wheels for flash-attn 2.8.3 on CUDA 12.8 + PyTorch 2.11 with cxx11abiTRUE.

Assets

• Python 3.10: flash_attn-2.8.3+cu12torch2.11cxx11abiTRUE-cp310-cp310-linux_x86_64.whl
• Python 3.11: flash_attn-2.8.3+cu12torch2.11cxx11abiTRUE-cp311-cp311-linux_x86_64.whl
• Python 3.12: flash_attn-2.8.3+cu12torch2.11cxx11abiTRUE-cp312-cp312-linux_x86_64.whl
• Python 3.13: flash_attn-2.8.3+cu12torch2.11cxx11abiTRUE-cp313-cp313-linux_x86_64.whl

SHA256

• flash_attn-2.8.3+cu12torch2.11cxx11abiTRUE-cp310-cp310-linux_x86_64.whl: 9001c730642cdc1ea44ed8130b0dc80e763519d6efc01e4de44b0700a0dfa13d
• flash_attn-2.8.3+cu12torch2.11cxx11abiTRUE-cp311-cp311-linux_x86_64.whl: cc09fc5be4a0005c93a37792ccfadf0ae544a53fdbc6a505649c3572849b8eb3
• flash_attn-2.8.3+cu12torch2.11cxx11abiTRUE-cp312-cp312-linux_x86_64.whl: 3d0c8e60f820321eedd7166e79c33cb816263d8be6e35c3f5ba8fe2df6fea697
• flash_attn-2.8.3+cu12torch2.11cxx11abiTRUE-cp313-cp313-linux_x86_64.whl: e0e889d708ae8a2af2b8a6cd6f6b45a8c32182cee68940685797f1cc239bef39

Build environment

flash-attention	v2.8.3
CUDA	12.8
PyTorch	2.11 (stable)
Python	3.10, 3.11, 3.12, 3.13
CXX11 ABI	TRUE
OS	Ubuntu 24.04

Notes

• Asset naming follows flash_attn-2.8.3+cu12torch2.11cxx11abiTRUE-....
• Python 3.9 is not included because upstream torch 2.11.0+cu128 does not publish Linux x86_64 cp39 wheels.