SageAttention2++: A More Efficient Implementation of SageAttention2

TL;DR

SageAttention2++ introduces a more efficient implementation of attention by leveraging FP8 Matmul instructions, achieving significant speedups while maintaining accuracy across multiple model types.

Contribution

It proposes using FP8 Matmul with FP16 accumulation to accelerate SageAttention2, resulting in nearly 4x speedup over FlashAttention without accuracy loss.

Findings

SageAttention2++ is 3.9x faster than FlashAttention.

Maintains the same attention accuracy as SageAttention2.

Effective across language, image, and video models.

Abstract

The efficiency of attention is critical because its time complexity grows quadratically with sequence length. SageAttention2 addresses this by utilizing quantization to accelerate matrix multiplications (Matmul) in attention. To further accelerate SageAttention2, we propose to utilize the faster instruction of FP8 Matmul accumulated in FP16. The instruction is 2x faster than the FP8 Matmul used in SageAttention2. Our experiments show that SageAttention2++ achieves a 3.9x speedup over FlashAttention while maintaining the same attention accuracy as SageAttention2. This means SageAttention2++ effectively accelerates various models, including those for language, image, and video generation, with negligible end-to-end metrics loss. The code will be available at https://github.com/thu-ml/SageAttention.