From 51b67595f4b1c57f48b679de3c20d9c19a764354 Mon Sep 17 00:00:00 2001 From: turboderp <11859846+turboderp@users.noreply.github.com> Date: Sun, 3 May 2026 00:01:29 +0200 Subject: [PATCH] Dependencies: Switch to mjun0812 flash-attn wheels --- pyproject.toml | 20 ++++++++++---------- 1 file changed, 10 insertions(+), 10 deletions(-) diff --git a/pyproject.toml b/pyproject.toml index 663e1bf..17dc8c0 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -88,17 +88,17 @@ cu12 = [ "exllamav3 @ https://github.com/turboderp-org/exllamav3/releases/download/v0.0.31/exllamav3-0.0.31+cu128.torch2.9.0-cp311-cp311-linux_x86_64.whl ; platform_system == 'Linux' and platform_machine == 'x86_64' and python_version == '3.11'", "exllamav3 @ https://github.com/turboderp-org/exllamav3/releases/download/v0.0.31/exllamav3-0.0.31+cu128.torch2.9.0-cp310-cp310-linux_x86_64.whl ; platform_system == 'Linux' and platform_machine == 'x86_64' and python_version == '3.10'", - # Windows FA2 from https://github.com/kingbri1/flash-attention/releases - "flash_attn @ https://github.com/kingbri1/flash-attention/releases/download/v2.8.3/flash_attn-2.8.3+cu128torch2.9.0cxx11abiFALSE-cp313-cp313-win_amd64.whl ; platform_system == 'Windows' and python_version == '3.13'", - "flash_attn @ https://github.com/kingbri1/flash-attention/releases/download/v2.8.3/flash_attn-2.8.3+cu128torch2.9.0cxx11abiFALSE-cp312-cp312-win_amd64.whl ; platform_system == 'Windows' and python_version == '3.12'", - "flash_attn @ https://github.com/kingbri1/flash-attention/releases/download/v2.8.3/flash_attn-2.8.3+cu128torch2.9.0cxx11abiFALSE-cp311-cp311-win_amd64.whl ; platform_system == 'Windows' and python_version == '3.11'", - "flash_attn @ https://github.com/kingbri1/flash-attention/releases/download/v2.8.3/flash_attn-2.8.3+cu128torch2.9.0cxx11abiFALSE-cp310-cp310-win_amd64.whl ; platform_system == 'Windows' and python_version == '3.10'", +# Windows FA2 from https://github.com/mjun0812/flash-attention-prebuild-wheels/releases +"flash_attn @ https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.7.13/flash_attn-2.8.3+cu128torch2.9-cp313-cp313-win_amd64.whl ; platform_system == 'Windows' and python_version == '3.13'", +"flash_attn @ https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.7.13/flash_attn-2.8.3+cu128torch2.9-cp312-cp312-win_amd64.whl ; platform_system == 'Windows' and python_version == '3.12'", +"flash_attn @ https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.7.13/flash_attn-2.8.3+cu128torch2.9-cp311-cp311-win_amd64.whl ; platform_system == 'Windows' and python_version == '3.11'", +"flash_attn @ https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.7.13/flash_attn-2.8.3+cu128torch2.9-cp310-cp310-win_amd64.whl ; platform_system == 'Windows' and python_version == '3.10'", - # Linux FA2 from https://github.com/kingbri1/flash-attention/releases - "flash_attn @ https://github.com/kingbri1/flash-attention/releases/download/v2.8.3/flash_attn-2.8.3+cu128torch2.9.0cxx11abiFALSE-cp313-cp313-linux_x86_64.whl ; platform_system == 'Linux' and platform_machine == 'x86_64' and python_version == '3.13'", - "flash_attn @ https://github.com/kingbri1/flash-attention/releases/download/v2.8.3/flash_attn-2.8.3+cu128torch2.9.0cxx11abiFALSE-cp312-cp312-linux_x86_64.whl ; platform_system == 'Linux' and platform_machine == 'x86_64' and python_version == '3.12'", - "flash_attn @ https://github.com/kingbri1/flash-attention/releases/download/v2.8.3/flash_attn-2.8.3+cu128torch2.9.0cxx11abiFALSE-cp311-cp311-linux_x86_64.whl ; platform_system == 'Linux' and platform_machine == 'x86_64' and python_version == '3.11'", - "flash_attn @ https://github.com/kingbri1/flash-attention/releases/download/v2.8.3/flash_attn-2.8.3+cu128torch2.9.0cxx11abiFALSE-cp310-cp310-linux_x86_64.whl ; platform_system == 'Linux' and platform_machine == 'x86_64' and python_version == '3.10'", +# Linux FA2 from https://github.com/mjun0812/flash-attention-prebuild-wheels/releases +"flash_attn @ https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.9.0/flash_attn-2.8.3+cu128torch2.9-cp313-cp313-linux_x86_64.whl ; platform_system == 'Linux' and platform_machine == 'x86_64' and python_version == '3.13'", +"flash_attn @ https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.9.0/flash_attn-2.8.3+cu128torch2.9-cp312-cp312-linux_x86_64.whl ; platform_system == 'Linux' and platform_machine == 'x86_64' and python_version == '3.12'", +"flash_attn @ https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.9.0/flash_attn-2.8.3+cu128torch2.9-cp311-cp311-linux_x86_64.whl ; platform_system == 'Linux' and platform_machine == 'x86_64' and python_version == '3.11'", +"flash_attn @ https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.9.0/flash_attn-2.8.3+cu128torch2.9-cp310-cp310-linux_x86_64.whl ; platform_system == 'Linux' and platform_machine == 'x86_64' and python_version == '3.10'", ] amd = [ # Torch triton for ROCm