Fast Generation from Convolutional Sequence Models
Application
US20260080220A1
Kind: A1
Mar 19, 2026
Inventors
Elad Hazan, Naman Agarwal, Daniel Can Suo, Xinyi Chen, Evan Andrew Dogariu, Vladimir Iouri Feinberg
Abstract
Provided are fast auto-regressive generation of sequence prediction models. For sequence prediction models that are based on convolutional operators, such as spectral state space models, example implementations reduce generation time from linear in the context length to square root of the context length.
CPC Classifications
G06N 3/0464
Filing Date
2025-09-16
Application No.
19330724