Views
No views yet
sageattention-2.2.0+cu131torch2.12.blackwell-cp312-cp312-win_amd64.whl| Component | Version |
|---|---|
| GPU | NVIDIA RTX 50 series (sm_120 Blackwell) |
| OS | Windows 10/11 x64 |
| Python | 3.12 |
| PyTorch | 2.12.0 nightly cu128 |
| Triton | 3.6.0 |
1pip install torch==2.12.0.dev20260402+cu128 --index-url https://download.pytorch.org/whl/nightly/cu128
2pip install triton==3.6.0pip install sageattention-2.2.0+cu131torch2.12.blackwell-cp312-cp312-win_amd64.whl --no-depsImportant: Always use--no-depsto prevent pip from overwriting your PyTorch installation.
1import torch
2print(torch.__version__) # 2.12.0.dev...
3
4from sageattention import sageattn_varlen
5print("SageAttention 2.2.0 OK")| Resolution | DiT Time | Total | VRAM |
|---|---|---|---|
| 4K (2160p) | 2.59s | 5.7s | 19.1GB |
| 12MP (3000p) | 4.68s | 9.2s | 22.4GB |
1git clone https://github.com/thu-ml/SageAttention.git
2cd SageAttention
3
4# Open fresh CMD, activate MSVC first
5call "C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Auxiliary\Build\vcvars64.bat"
6
7set CUDA_HOME=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1
8set TORCH_CUDA_ARCH_LIST=12.0
9set DISTUTILS_USE_SDK=1
10set MAX_JOBS=4#ifdef __CUDACC__ guards to 3 PyTorch header files to prevent CUDA/C++ compilation conflicts:torch/include/torch/csrc/dynamo/compiled_autograd.h — fixes C2872 'std' ambiguoustorch/include/torch/csrc/autograd/custom_function.h — fixes incomplete PackedArgs typetorch/include/torch/csrc/autograd/_functions.h — fixes missing autograd::Functiontorch/torch.h and torch/cuda.h includes from SA source .cu and .cuh files (replace with lighter headers).pip install . --no-build-isolationImportError: DLL load failed. Rebuild from source if needed.vcvars64.bat adds ~2000 chars to PATH each call. Use a fresh CMD window to avoid "The input line is too long" errors.