New deepseek paper: Natively Trainable Sparse Attention mechanism twitter.com 5 points by redlock a year ago · 2 comments Reader PiP Save eunos a year ago Authored and Uploaded by none others than Liang Wenfeng himself