EDIT: Early Diffusion Inference Termination for dLLMs Based on Dynamics of Training Gradients

TL;DR

This paper introduces EDIT, a method that adaptively terminates diffusion-based large language model inference by monitoring training-gradient-derived reasoning stability, significantly reducing inference steps while maintaining accuracy.

Contribution

We propose a novel inference-time criterion using training-gradient dynamics to efficiently stop diffusion processes in dLLMs, reducing computation without sacrificing performance.

Findings

Reduces diffusion steps by up to 68.3%

Maintains or improves accuracy across reasoning benchmarks

Adds minimal storage overhead (~0.02%)

Abstract

Diffusion-based large language models (dLLMs) refine token generations through iterative denoising, but answers often stabilize before all steps complete. We propose EDIT (Early Diffusion Inference Termination), an inference-time criterion that adaptively stops denoising once sufficient reasoning stability relative to training-time reasoning is detected. EDIT monitors the alignment between token activations and a reasoning map derived from AdamW-aggregated LoRA updates captured during supervised fine-tuning (SFT). During training, optimization dynamics generate rich metadata about parameter importance that in prior methods is typically discarded upon model release. We preserve this information as a compact representation of learned reasoning pathways. During inference, alignment scores are converted to a distribution over the tokens already unmasked at the current denoising step, and convergence is detected when KL divergence between consecutive steps falls below a threshold on the matched unmasked (visible) tokens. Across reasoning benchmarks, EDIT reduces diffusion steps by 11.8% to 68.3% while preserving or improving accuracy in most settings, with approximately 0.02% storage overhead (about 1.5-2 MB for all QKV modules across 32 blocks in an 8 GB model). By utilizing training-gradient dynamics, our work opens a new research direction for reducing dLLM inference time and cost.