KLASS: KL-Guided Fast Inference in Masked Diffusion Models
This addresses the inference bottleneck in masked diffusion models for tasks like language generation, offering a broadly applicable solution with significant speed improvements.
The paper tackles the slow inference speed of masked diffusion models by introducing KLASS, a sampling method that uses token-level KL divergence to unmask multiple tokens per iteration, achieving up to 2.78x speedups while maintaining or improving sample quality across various domains.
Masked diffusion models have demonstrated competitive results on various tasks including language generation. However, due to its iterative refinement process, the inference is often bottlenecked by slow and static sampling speed. To overcome this problem, we introduce `KL-Adaptive Stability Sampling' (KLASS), a fast yet effective sampling method that exploits token-level KL divergence to identify stable, high-confidence predictions. By unmasking multiple tokens in each iteration without any additional model training, our approach speeds up generation significantly while maintaining sample quality. On reasoning benchmarks, KLASS achieves up to $2.78\times$ wall-clock speedups while improving performance over standard greedy decoding, attaining state-of-the-art results among diffusion-based samplers. We further validate KLASS across diverse domains, including text, image, and molecular generation, showing its effectiveness as a broadly applicable sampler across different models.