Accelerating Diffusion LLM Inference via Local Determinism Propagation
This addresses inference inefficiencies for deploying dLLMs in practical applications, representing an incremental improvement over existing methods.
The paper tackled the quality-speed trade-off in diffusion large language models (dLLMs) by proposing LocalLeap, a training-free adaptive parallel decoding strategy that reduces decoding steps to 14.2% of the original and achieves 6.94× throughput improvements with negligible performance impact.
Diffusion large language models (dLLMs) represent a significant advancement in text generation, offering parallel token decoding capabilities. However, existing open-source implementations suffer from quality-speed trade-offs that impede their practical deployment. Conservative sampling strategies typically decode only the most confident token per step to ensure quality (i.e., greedy decoding), at the cost of inference efficiency due to repeated redundant refinement iterations--a phenomenon we term delayed decoding. Through systematic analysis of dLLM decoding dynamics, we characterize this delayed decoding behavior and propose a training-free adaptive parallel decoding strategy, named LocalLeap, to address these inefficiencies. LocalLeap is built on two fundamental empirical principles: local determinism propagation centered on high-confidence anchors and progressive spatial consistency decay. By applying these principles, LocalLeap identifies anchors and performs localized relaxed parallel decoding within bounded neighborhoods, achieving substantial inference step reduction through early commitment of already-determined tokens without compromising output quality. Comprehensive evaluation on various benchmarks demonstrates that LocalLeap achieves 6.94$\times$ throughput improvements and reduces decoding steps to just 14.2\% of the original requirement, achieving these gains with negligible performance impact. The source codes are available at: https://github.com/friedrichor/LocalLeap.