Paper 2602.06036
DFlash: Block Diffusion for Flash Speculative Decoding
- Published
- Feb 2026
- Research lab
- Independent
- Citations
- 63
- GitHub
- 5.6K stars
01 In brief
Summary
DFlash is a speculative decoding framework that uses a lightweight block diffusion model for parallel drafting, addressing the sequential bottleneck of autoregressive LLM decoding.
It extracts hidden features from the target LLM and injects them into every draft layer's KV cache, conditioning the draft model to predict future token blocks in parallel.
This design enables high acceptance rates and low drafting latency, achieving over 6x lossless acceleration across models and tasks, and up to 2.5x higher speedup than the state-of-the-art EAGLE-3.
Key training innovations include random sampling of masked blocks, exponentially decaying loss weights for early tokens, and shared embeddings with the target model.
Ablations show that KV injection outperforms input fusion, and models trained with larger block sizes generalize well to smaller inference-time block sizes.
DFlash also demonstrates practical speedups in serving frameworks like SGLang and vLLM, and adapts to long contexts with minimal fine-tuning.
02 From the paper
Abstract
Autoregressive large language models (LLMs) deliver strong performance but require inherently sequential decoding, leading to high inference latency and poor GPU utilization. Speculative decoding mitigates this bottleneck by using a fast draft model whose outputs are verified in parallel by the target LLM; however, existing methods still rely on autoregressive drafting, which remains sequential and limits practical speedups. Diffusion LLMs offer a promising alternative by enabling parallel generation, but current diffusion models typically underperform compared with autoregressive models. In this paper, we introduce DFlash, a speculative decoding framework that employs a lightweight block diffusion model for parallel drafting. By generating draft tokens in a single forward pass and conditioning the draft model on context features extracted from the target model, DFlash enables efficient drafting with high-quality outputs and higher acceptance rates. Experiments show that DFlash achieves over 6x lossless acceleration across a range of models and tasks, delivering up to 2.5x higher speedup than the state-of-the-art speculative decoding method EAGLE-3.