Search papers, labs, and topics across Lattice.
3
0
4
2
Learning to prune key-value caches based on future token utility allows KVpop to achieve up to 88% compression while retaining nearly full performance.
xLSTM outperforms its competitors in complex sequence modeling tasks by leveraging advanced state tracking and memory correction mechanisms.
xLSTM models can now effectively learn from large attention-based models, even outperforming their teachers on some tasks through a novel distillation and merging pipeline.
