AdaFlash: Adaptive Speculative Decoding via On-Policy Distilled Diffusion Drafters

• URL: http://arxiv.org/abs/2607.19223v1
• Date: Tue, 21 Jul 2026 15:52:50 GMT
• Title: AdaFlash: Adaptive Speculative Decoding via On-Policy Distilled Diffusion Drafters
• Abstract summary: Speculative decoding has become a prevalent paradigm for accelerating large language model inference.<n>Recent work such as DFlash further boosts drafting efficiency by leveraging diffusion drafters.<n>We propose AdaFlash framework, comprising two components: (i) an on-policy distillation (OPD) algorithm with reverse-KL divergence tailored for diffusion drafters, and (ii) an adaptive length head that dynamically adjusts the candidate sequence length on the fly.
• Score: 67.51886052453452
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Speculative decoding, in which a lightweight draft model first generates a draft sequence that is then verified in parallel by the target model, has become a prevalent paradigm for accelerating large language model inference. Recent work such as DFlash further boosts drafting efficiency by leveraging diffusion drafters, whose parallel denoising mechanism enables draft generation in a single forward pass. In this work, we uncover a central pitfall of diffusion drafters: bidirectional attention is a double-edged sword. On one hand, it endows the model with parallel generation and global contextual modeling capabilities; on the other hand, this inherent global dependency introduces high variance at both the domain-level and the token-level: acceptance rates fluctuate substantially across different domains, and draft token quality also varies heterogeneously at different token positions. To tackle this issue, we propose AdaFlash framework, comprising two components: (i) an on-policy distillation (OPD) algorithm with reverse-KL divergence tailored for diffusion drafters, bringing stable convergence and effectively reducing domain-level variance; and (ii) an adaptive length head that dynamically adjusts the candidate sequence length on the fly, substantially lowering the verification cost of the target model and handling token-level variance. Experiments demonstrate that AdaFlash consistently improves speedup rate during deployment, with especially significant gains in high-concurrency scenarios, achieving up to approximately 66% higher throughput than previous state-of-the-art methods.

Related papers

• WAM-Diff2: Hierarchical AR-to-Diffusion Distillation for Highly Efficient Autonomous Driving VLA [27.102605117827398]
  Vision-Language-Action (VLA) models have emerged as a prominent paradigm for end-to-end autonomous driving.<n>They are constrained by high computational latency and exposure bias arising from sequential autoregressive decoding.<n>We introduce WAM-Diff2, a multi-task discrete diffusion VLA framework powered by a three-stage hierarchical strategy.
  arXiv  Detail & Related papers  (2026-08-02T06:45:16Z)
• WhiFlash: Accelerating Speculative Decoding with Token-Level Cross-Paradigm Routing [44.656073829954636]
  WhiFlash is a cross-paradigm SD method that unifies autoregressive and diffusion-based parallel drafting under a single token-level controller.<n>We show that WhiFlash achieves significantly higher acceptance lengths, yielding category-specific throughput gains of up to 69.6%.
  arXiv  Detail & Related papers  (2026-06-05T13:23:33Z)
• Diffusion-CAM: Faithful Visual Explanations for dMLLMs [8.78213424930081]
  Diffusion-CAM is the first interpretability method specifically tailored for dMLLMs.<n>We derive raw activation maps by differentiably probing intermediate representations in the backbone.<n>Our experiments demonstrate that Diffusion-CAM significantly outperforms SoTA methods in both localization accuracy and visual fidelity.
  arXiv  Detail & Related papers  (2026-04-13T05:14:18Z)
• OmniVL-Guard: Towards Unified Vision-Language Forgery Detection and Grounding via Balanced RL [63.388513841293616]
  Existing forgery detection methods fail to handle the interleaved text, images, and videos prevalent in real-world misinformation.<n>To bridge this gap, this paper targets to develop a unified framework for omnibus vision-language forgery detection and grounding.<n>We propose textbf OmniVL-Guard, a balanced reinforcement learning framework for omnibus vision-language forgery detection and grounding.
  arXiv  Detail & Related papers  (2026-02-11T09:41:36Z)
• Beyond Confidence: Adaptive and Coherent Decoding for Diffusion Language Models [64.92045568376705]
  Coherent Contextual Decoding (CCD) is a novel inference framework built upon two core innovations.<n>CCD employs a trajectory rectification mechanism that leverages historical context to enhance sequence coherence.<n>Instead of rigid allocations based on diffusion steps, we introduce an adaptive sampling strategy that dynamically adjusts the unmasking budget for each step.
  arXiv  Detail & Related papers  (2025-11-26T09:49:48Z)
• Dual-Stream Diffusion for World-Model Augmented Vision-Language-Action Model [62.889356203346985]
  We propose DUal-STream diffusion (DUST), a world-model augmented VLA framework that handles the modality conflict.<n>DUST achieves up to 6% gains over a standard VLA baseline and implicit world-modeling methods.<n>On real-world tasks with the Franka Research 3, DUST outperforms baselines in success rate by 13%.
  arXiv  Detail & Related papers  (2025-10-31T16:32:12Z)
• Next Interest Flow: A Generative Pre-training Paradigm for Recommender Systems by Modeling All-domain Movelines [8.895768051554162]
  We propose a novel generative pre-training paradigm for e-commerce recommender systems.<n>Our model learns to predict the Next Interest Flow, a dense vector sequence representing a user's future intent.<n>We present the All-domain Moveline Evolution Network (AMEN), a unified framework implementing our entire pipeline.
  arXiv  Detail & Related papers  (2025-10-13T12:13:17Z)
• SPREAD: Sampling-based Pareto front Refinement via Efficient Adaptive Diffusion [0.8594140167290097]
  SPREAD is a generative framework based on Denoising Diffusion Probabilistic Models (DDPMs)<n>It learns a conditional diffusion process over points sampled from the decision space.<n>It refines candidates via a sampling scheme that uses an adaptive multiple gradient descent-inspired update for fast convergence.
  arXiv  Detail & Related papers  (2025-09-25T12:09:37Z)
• Multimodal LLM-Guided Semantic Correction in Text-to-Image Diffusion [52.315729095824906]
  MLLM Semantic-Corrected Ping-Pong-Ahead Diffusion (PPAD) is a novel framework that introduces a Multimodal Large Language Model (MLLM) as a semantic observer during inference.<n>It performs real-time analysis on intermediate generations, identifies latent semantic inconsistencies, and translates feedback into controllable signals that actively guide the remaining denoising steps.<n>Extensive experiments demonstrate PPAD's significant improvements.
  arXiv  Detail & Related papers  (2025-05-26T14:42:35Z)
• One-Step Diffusion Model for Image Motion-Deblurring [85.76149042561507]
  We propose a one-step diffusion model for deblurring (OSDD), a novel framework that reduces the denoising process to a single step.<n>To tackle fidelity loss in diffusion models, we introduce an enhanced variational autoencoder (eVAE), which improves structural restoration.<n>Our method achieves strong performance on both full and no-reference metrics.
  arXiv  Detail & Related papers  (2025-03-09T09:39:57Z)
• Speculative Diffusion Decoding: Accelerating Language Generation through Diffusion [55.0194604505437]
  Speculative decoding has emerged as a widely adopted method to accelerate large language model inference.<n>This paper proposes an adaptation of speculative decoding which uses discrete diffusion models to generate draft sequences.
  arXiv  Detail & Related papers  (2024-08-10T21:24:25Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.