---
library_name: kernels
license: apache-2.0
---

This is the repository card of yetter-ai/sage-attention that has been pushed on the Hub. It was built to be used with the [`kernels` library](https://github.com/huggingface/kernels).

Build source: https://github.com/thu-ml/SageAttention
Upstream commit: d1a57a546c3d395b1ffcbeecc66d81db76f3b4b5

## How to use

```python
# make sure `kernels` is installed: `pip install -U kernels`
from kernels import get_kernel

kernel_module = get_kernel("yetter-ai/sage-attention")
sageattn = kernel_module.sageattn

sageattn(...)
```

## Available functions
- `per_block_int8`
- `per_warp_int8`
- `sub_mean`
- `per_channel_fp8`
- `sageattn`
- `sageattn_varlen`
- `sageattn_qk_int8_pv_fp16_triton`
- `sageattn_qk_int8_pv_fp16_cuda`
- `sageattn_qk_int8_pv_fp8_cuda`
- `sageattn_qk_int8_pv_fp8_cuda_sm90`
- `sageattn3_blackwell`

## Benchmarks

No benchmark available yet.