New deepseek paper: Natively Trainable Sparse Attention mechanism twitter.com 5 points by redlock 10 months ago · 2 comments Reader PiP Save eunos 10 months ago Authored and Uploaded by none others than Liang Wenfeng himself