Search papers, labs, and topics across Lattice.
Affiliation:
1
0
2
5
Achieve 4-bit quantization for diffusion LLMs without calibration data, slashing memory by 2.6x and inference by 2.2x while preserving accuracy.
