FlashDLM: Accelerating Diffusion Language Model Inference via Efficient KV Caching and Guided Diffusion

TL;DR

This paper introduces FlashDLM, a method that significantly accelerates diffusion language model inference by using KV caching and guided diffusion, achieving comparable or faster speeds than autoregressive models with minimal quality loss.

Contribution

The paper presents two training-free techniques, FreeCache and Guided Diffusion, to reduce inference time and complexity of diffusion language models without sacrificing quality.

Findings

Achieves 12.14x speedup in inference

Diffusion models match or surpass AR model latency

Minimal accuracy degradation observed

Abstract

Diffusion language models offer parallel token generation and inherent bidirectionality, promising more efficient and powerful sequence modeling compared to autoregressive approaches. However, state-of-the-art diffusion models (e.g., Dream 7B, LLaDA 8B) suffer from slow inference. While they match the quality of similarly sized autoregressive (AR) models (e.g., Qwen2.5 7B, Llama3 8B), their iterative denoising requires multiple full-sequence forward passes, resulting in high computational costs and latency, particularly for long input prompts and long-context scenarios. Furthermore, parallel token generation introduces token incoherence problems, and current sampling heuristics suffer from significant quality drops with decreasing denoising steps. We address these limitations with two training-free techniques. First, we propose FreeCache, a Key-Value (KV) approximation caching technique that reuses stable KV projections across denoising steps, effectively reducing the computational cost of DLM inference. Second, we introduce Guided Diffusion, a training-free method that uses a lightweight pretrained autoregressive model to supervise token unmasking, dramatically reducing the total number of denoising iterations without sacrificing quality. We conduct extensive evaluations on open-source reasoning benchmarks, and our combined methods deliver an average of 12.14x end-to-end speedup across various tasks with negligible accuracy degradation. For the first time, diffusion language models achieve a comparable and even faster latency as the widely adopted autoregressive models. Our work successfully paved the way for scaling up the diffusion language model to a broader scope of applications across different domains.