T twitter.com New deepseek paper: Natively Trainable Sparse Attention mechanism 5 points redlock 1 year ago 1 comments
