Optimizing Few-Step Generation with Adaptive Matching Distillation

2026-02-07

AI Summary

This work addresses the instability in distribution matching distillation (DMD) caused by unreliable teacher signals within "forbidden regions," which often leads to training collapse. To resolve this, we propose Adaptive Matching Distillation (AMD), a novel framework that explicitly models and corrects optimization trajectories in these problematic regions. AMD employs a reward agent to dynamically detect and escape forbidden zones, integrating structural signal decomposition with a repulsive landscape sharpening mechanism. This adaptively reweights gradient priorities and reinforces energy barriers to prevent mode collapse. Experiments on models such as SDXL and Wan2.1 demonstrate significant improvements in few-step generation performance, with the HPSv2 score on SDXL rising from 30.64 to 31.25. Further evaluations on VBench and GenEval confirm concurrent gains in generation fidelity and training robustness.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Distributed Machine Learning & Federated Learning
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Vertical and domain-specific search

Abstract

Distribution Matching Distillation (DMD) is a powerful acceleration paradigm, yet its stability is often compromised in Forbidden Zone, regions where the real teacher provides unreliable guidance while the fake teacher exerts insufficient repulsive force. In this work, we propose a unified optimization framework that reinterprets prior art as implicit strategies to avoid these corrupted regions. Based on this insight, we introduce Adaptive Matching Distillation (AMD), a self-correcting mechanism that utilizes reward proxies to explicitly detect and escape Forbidden Zones. AMD dynamically prioritizes corrective gradients via structural signal decomposition and introduces Repulsive Landscape Sharpening to enforce steep energy barriers against failure mode collapse. Extensive experiments across image and video generation tasks (e.g., SDXL, Wan2.1) and rigorous benchmarks (e.g., VBench, GenEval) demonstrate that AMD significantly enhances sample fidelity and training robustness. For instance, AMD improves the HPSv2 score on SDXL from 30.64 to 31.25, outperforming state-of-the-art baselines. These findings validate that explicitly rectifying optimization trajectories within Forbidden Zones is essential for pushing the performance ceiling of few-step generative models.

Problem

• Few-Step Generation
• Distribution Matching Distillation
• Forbidden Zone
• Optimization Stability
• Generative Models

Innovation

• Adaptive Matching Distillation
• Forbidden Zone
• Repulsive Landscape Sharpening
• Few-Step Generation
• Distribution Matching Distillation