New deepseek paper: Natively Trainable Sparse Attention mechanism (twitter.com) 5 points by redlock 1y ago ↗ HN
2 comments
[ 2.9 ms ] story [ 16.1 ms ] thread