目录
Background
参考
Flash Attention: Fast and Memory-Efficient Exact Attention with IO-Awareness
Background
HBM: high Boardwidth memory, 高带宽内存
SRAM: Static RAM, 静态随机存储器
Flash Attention 分而治之的思想
参考
Flash Attention论文解读 - 李理的博客
https://gordicaleksa.medium.com/eli5-flash-attention-5c44017022ad