{"ID":2847290,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2511.00606","arxiv_id":"2511.00606","title":"SpecDiff-2: Scaling Diffusion Drafter Alignment For Faster Speculative Decoding","abstract":"Speculative decoding has become the standard approach for accelerating Large Language Model (LLM) inference. It exploits a lossless draft-then-verify procedure to circumvent the latency of autoregressive decoding, achieving impressive speed-ups. Yet, current speculative decoding approaches remain limited by two fundamental bottlenecks: (1) the autoregressive dependency during drafting which limits parallelism, and (2) frequent rejections of draft tokens caused by misalignment between the draft and verify models. This paper proposes SpecDiff-2, a novel framework to jointly address these two bottlenecks. It leverages discrete diffusion as a non-autoregressive drafter to address bottleneck (1) and develops novel techniques to calibrate discrete diffusion drafters with autoregressive verifiers, addressing bottleneck (2). Experimental results across a comprehensive benchmark suite show that SpecDiff-2 achieves a new state-of-the-art across reasoning, coding, and mathematical benchmarks, improving tokens-per-second by up to an average of +55% over previous baselines and obtaining up to 5.5x average speed-up over standard decoding, without any loss of accuracy.","short_abstract":"Speculative decoding has become the standard approach for accelerating Large Language Model (LLM) inference. It exploits a lossless draft-then-verify procedure to circumvent the latency of autoregressive decoding, achieving impressive speed-ups. Yet, current speculative decoding approaches remain limited by two fundame...","url_abs":"https://arxiv.org/abs/2511.00606","url_pdf":"https://arxiv.org/pdf/2511.00606v2","authors":"[\"Jameson Sandler\",\"Jacob K. Christopher\",\"Thomas Hartvigsen\",\"Ferdinando Fioretto\"]","published":"2025-11-01T16:12:56Z","proceeding":"cs.CL","tasks":"[\"cs.CL\"]","methods":"[\"Diffusion Model\",\"Large Language Model\",\"Language Model\"]","has_code":false}