Search papers, labs, and topics across Lattice.
Affiliation:
4
0
7
Achieving superior simultaneous speech translation quality without altering LLM architecture could redefine efficiency in real-time translation tasks.
MetaRouter learns user preferences for LLM routing with minimal interaction, outperforming traditional methods in diverse scenarios.
By aggregating Gram matrices instead of low-rank factors, FLoRG sidesteps decomposition drift in federated LoRA, achieving up to 2041x communication savings and improved accuracy.