Search papers, labs, and topics across Lattice.
University of California
3
0
4
Reading barely 20% of the KV cache is enough to match up to 99% of dense attention accuracy at 32K context, yielding an immediate 1.69× speedup over FlashAttention without any retraining.
A million p-bits in a single programmable architecture reveals a universal tradeoff between throughput and accuracy in distributed probabilistic computing.
Attention bottlenecks in long-context decoding? SANTA slashes memory bandwidth demands by stochastically sampling value vectors, achieving 1.5x speedups without sacrificing accuracy.
