Search papers, labs, and topics across Lattice.
MiniMax
1
0
2
10
MSA slashes per-token attention compute by over 28x while maintaining competitive performance, revolutionizing how LLMs can handle ultra-long contexts.
