AI
Aligning One-Step Generative Models with Reward-Weighted Transport Distillation
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 24/09/2026, 20:00:00 UTC
Briefing created: 02/10/2026, 10:26:53 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 01/10/2026, 19:18:24 UTC.
Briefings
This article presents Reward-Weighted Transport Distillation (RWTD), a post-training method for one-step generative models that improves GenEval scores from 0.73 to 0.80 by mixing tilted current and reference distributions. The method uses feature-space optimal transport and fixed-point regression to balance reward adaptation with prior knowledge, showing strong cross-reward generalization and compositional capability preservation.
Key points
- RWTD constructs an adaptive target by mixing separately tilted current and reference distributions using feature-space optimal transport and fixed-point regression.
- The method improves GenEval scores from 0.73 to 0.80 on the one-step SANA Sprint 1.6B backbone, demonstrating strong cross-reward generalization.
Context and limitations
The study relies on generated samples and scalar reward evaluations; theoretical analysis is limited to fixed-point distributions, and empirical results are specific to the SANA Sprint 1.6B backbone.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 02/10/2026, 10:26:53 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeiaskjmsqt3ad6nvyzna2ptvw7fnvigifqztfbdsxcu7wlhqmb5hxu