Search papers, labs, and topics across Lattice.
State Key Lab of IOTSC, University of Macau
6
0
10
Reorganizing expert groups in MoE models can slash accuracy degradation by over 71% while boosting throughput by more than twofold.
Cold start latencies can be slashed by up to 99.3% with CoCoScale's innovative layer-wise scaling approach.
SwiftCache slashes latency by 69% and boosts context length nearly fourfold by enabling cross-model KV cache sharing on GPUs.
Injecting image-derived semantics into LLMs dramatically improves their ability to detect AI-generated Chinese poetry, surpassing even specialized text-based detectors.
Forget optimizing for sensing or compute – MA-EQA performance skyrockets when you allocate communication power based on how well agents *remember* what they've seen.
LLM scaling bottlenecks demand a shift towards cloud-native architectures and distributed systems, unlocking potential gains from serverless inference and quantum computing.
