Search papers, labs, and topics across Lattice.
2
0
5
0
A frozen 12B model achieves 100% accuracy on new problem instances at zero generation tokens, challenging the need for constant retraining in language models.
A frozen 12B model can leap from 80% to 93.3% accuracy by leveraging byte-exact KV-cache grafting, achieving unprecedented efficiency with 8,700x less energy consumption.
