Get the latest tech news
FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-Precision
precision Attention, as a core layer of the ubiquitous Transformer architecture, is a bottleneck for large language models and long-context applications. FlashAttention (and FlashAttention-2) pioneered an approach to speed up attention on GPUs by minimizing memory reads/writes, and is now used by most libraries to accelerate Transformer training and inference.
In this blogpost, we describe three main techniques to speed up attention on Hopper GPUs: exploiting asynchrony of the Tensor Cores and TMA to (1) overlap overall computation and data movement via warp-specialization and (2) interleave block-wise matmul and softmax operations, and (3) incoherent processing that leverages hardware support for FP8 low-precision. FlashAttention is an algorithm that reorders the attention computation and leverages tiling and recomputation to significantly speed it up and reduce memory usage from quadratic to linear in sequence length. In our experiment where Q, K, V are generated from a standard normal distribution but 0.1% of the entries have large magnitudes (to simulate outliers), we found that incoherent processing can reduce the quantization error by 2.6x.
Or read this on Hacker News
