Search papers, labs, and topics across Lattice.
2
0
5
Forget outcome-based filtering: TopoCurate uses interaction topology to surface informative tool-use trajectories and tasks, boosting SFT and RL performance by up to 6.9%.
Multi-Head Low-Rank Attention (MLRA) unlocks 2.8x faster distributed decoding by enabling partitionable latent states, overcoming the sharding bottleneck of previous latent attention methods.
