A one-step generation model with a Single-Layer Transformer: Layer number re-distillation of FreeFlow

Haonan Wei, Linyuan Wang, Nuolin Sun, Zhizhong Zheng, Lei Li, Bin Yan · 2026

Read A one-step generation model with a Single-Layer Transformer: Layer number re-distillation of FreeFlow by Haonan Wei, Linyuan Wang, Nuolin Sun, Zhizhong Zheng, Lei Li, Bin Yan (2026) 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-2601-11630.txt.

Description

Haonan Wei and five coauthors distill a generation model to one layer. Their 2026 preprint reworks FreeFlow, a one-step flow method. Flow matching compresses diffusion generation into few steps. MeanFlow and FreeFlow lead that one-step push via ODEs.

FreeFlow uses a 28-layer transformer for its generation. The team reads those layers as steps of an Euler scheme. Depth acts as discrete time along an ODE path. That view suggests the layers can be re-distilled. Their model generates in one step with a single-layer transformer. A one-layer generator cuts cost and latency sharply. The work tests how much depth generation truly needs.

Catalog record

CREATORHaonan Wei, Linyuan Wang, Nuolin Sun, Zhizhong Zheng, Lei Li, Bin Yan
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE23 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-11630

Browse the full library · About the Tower of Babel