MiniMax teased M3 Sparse Attention: 9.7x prefilling, 15.6x decoding at 1M

(twitter.com)

8 points | by rebekkamikkoa 9 hours ago ago

No comments yet.