Single-Step Consistent Diffusion Samplers

Pascal Jutras-Dubé, Patrick Pynadath, Ruqi Zhang · 2025

Read Single-Step Consistent Diffusion Samplers by Pascal Jutras-Dubé, Patrick Pynadath, Ruqi Zhang (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-2502-07579.txt.

Description

Jutras-Dube, Pynadath, and Zhang introduce in 2025 samplers that draw high grade samples from a target spread in a single step. Standard samplers need many rounds, which costs time and compute. Their class builds on diffusion models but skips the long chain.

One training path distills a sampler from a pre trained diffusion model, using partial noisy paths rather than huge pre made sample sets. A second path trains from scratch with a self consistency loss, so one model both samples and skips steps. Tests on varied target spreads gave strong samples with under one percent of the network calls of standard diffusion samplers.

Catalog record

CREATORPascal Jutras-Dubé, Patrick Pynadath, Ruqi Zhang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE44 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-07579

Browse the full library · About the Tower of Babel