Search papers, labs, and topics across Lattice.
Affiliation:
3
0
7
Selective re-scanning in recurrent networks can drastically reduce memory usage while improving task performance, challenging the conventional wisdom of fixed-size state fidelity.
TPUs aren't just for Google anymore: Gemma 4 fine-tuning is 1.6x faster and 2x cheaper than on GPUs, with faster time-to-first-token for inference.
LLMs trained with Vector Policy Optimization (VPO) learn to produce diverse solutions that unlock previously unsolvable problems in evolutionary search, outperforming models optimized for single scalar rewards.
