3 ms·
FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness
- ftxbro 4y agoThis is part of the architecture of MosaicML (https://news.ycombinator.com/item?id=35829800 https://news.ycombinator.com/item?id=35829800).
loading thread…