Search papers, labs, and topics across Lattice.
1
0
2
Achieve state-of-the-art 4-bit LLM inference without calibration data by deriving activation scales directly from weight reconstruction error.