AI Tools & Models HOTFeaturedAll
arXiv cs.AIScore60

MeanFlowAdvantage: Stable Reward Fine-Tuning for Few-Step Average-Velocity Generators

Summary

This work fixes the mismatch between reward fine-tuning and the average-velocity representation in few-step MeanFlow generation, stabilizing the tuning.

Why this matters

Relevant to this topic

Open original ↗

Source: arXiv cs.AI · arxiv.org

Related items