LANTERN++: Enhancing Relaxed Speculative Decoding with Static Tree Drafting for Visual Auto-regressive Models

Sihwan Park, Doohyuk Jang, Sungyub Kim, Souvik Kundu, Eunho Yang · 2025

Read LANTERN++: Enhancing Relaxed Speculative Decoding with Static Tree Drafting for Visual Auto-regressive Models by Sihwan Park, Doohyuk Jang, Sungyub Kim, Souvik Kundu, Eunho Yang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-06352.txt.

Description

Sihwan Park, Doohyuk Jang, Sungyub Kim, Souvik Kundu, and Eunho Yang speed image chain makers with set draft trees in a 2025 paper. Guess-and-check drafts speed text birth, yet image chains balk at the trick. Near-tied picks at each step stall the drafts and cut gains hard in tests.

A past loose form let draft trees grow on the fly with soft checks. Flat ties still kept those trees thin, the team finds in their runs. Their set trees plus a fit check let drafts stand free of weak picks.

Catalog record

CREATORSihwan Park, Doohyuk Jang, Sungyub Kim, Souvik Kundu, Eunho Yang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE39 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-06352

Browse the full library · About the Tower of Babel