RL Post-Training in Diffusion Transformers Is Activation Steering

Media reports Hugging Face Papers Source language: Chinese

A fixed set of steering vectors added to a frozen diffusion transformer recovers most quality gains from reinforcement learning post-training. This indicates that improvements stem largely from a constant shift rather than prompt-specific transformations.

Full text

A fixed set of steering vectors added to a frozen diffusion transformer recovers most of the quality gains from reinforcement learning post-training, indicating that much of the improvement comes from a constant shift rather than prompt-specific transformations.

Sources

← Back to Daily AI Report