Search papers, labs, and topics across Lattice.
6
0
10
Generative world action models can run up to 3× faster in closed-loop control without fine-tuning or task degradation by dynamically repairing and reusing intermediate computational states instead of recomputing them from scratch.
SAC achieves a staggering 2.1x throughput increase by fetching only the essential KV entries for sparse attention models, revolutionizing memory efficiency in LLM inference.
ShotCrop$^3$ transforms a single image into a powerful narrative tool by generating three distinct shots, outperforming existing models in shot localization accuracy.
DisagFusion unlocks up to 20x higher throughput for diffusion model serving by intelligently splitting the workload across heterogeneous GPUs and dynamically adapting to workload shifts.
Unlock accurate predictions of complex nonlinear system dynamics from just one shot, slashing data needs for MEMS characterization.
