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

 (twitter.com)

8 points | by rebekkamikkoa 10 hours ago ago

No comments yet.