Quant VideoGen studies long-horizon auto-regressive video generation and identifies KV-cache memory as a key bottleneck. The paper proposes quantization techniques tailored to video redundancy to reduce memory cost while preserving quality.
Quant VideoGen
Haocheng Xi*, Shuo Yang*, Yilong Zhao, Chenfeng Xu, Muyang Li, Xiuyu Li, Ion Stoica, Song Han, Kurt Keutzer
|
Feb 1, 2026
