Decoupled Contrastive Decoding via Expert-Aligned Drafting

• URL: http://arxiv.org/abs/2608.12913v1
• Date: Thu, 13 Aug 2026 07:56:16 GMT
• Title: Decoupled Contrastive Decoding via Expert-Aligned Drafting
• Abstract summary: Contrastive Decoding (CD) improves generation quality, but its amateur-model pass makes decoding expensive.<n>Accelerating CD with speculative decoding raises a proposal-alignment question: should the contrastive signal shape the drafter, or should it remain only in verification?<n>We introduce Decoupled Contrastive Decoding (DCD), which drafts with an expert-aligned lightweight proposer and applies the amateur only in unchanged CD verification.
• Score: 8.38712879811618
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Contrastive Decoding (CD) improves generation quality, but its amateur-model pass makes decoding expensive. Accelerating CD with speculative decoding raises a proposal-alignment question: should the contrastive signal shape the drafter, or should it remain only in verification? We study this question in the lightweight feature-level drafter regime. Two controlled diagnostics, matched Cross-alpha training and an Approximate Dual-Drafter decomposition, give the same diagnosis: contrastive-aware drafting does not consistently improve over expert-aligned drafting because the contrastive correction is usually weaker than drafter error, and reconstruction can amplify that error. We introduce Decoupled Contrastive Decoding (DCD), which drafts with an expert-aligned lightweight proposer and applies the amateur only in unchanged CD verification. Standard speculative verification preserves the vanilla-CD output distribution. Across the main 8B settings, EAGLE3-based DCD achieves average greedy speedups of 1.65 to 1.95x over vanilla CD and reduces MMLU proposal-path latency by about 5 to 12x relative to amateur-coupled proposal paths.

Related papers

• D-cut: Adaptive Verification Depth Pruning for Batched Speculative Decoding [14.590812900433328]
  Speculative decoding accelerates large language model (LLM) inference without compromising output quality.<n>We present D-Cut, an adaptive pruning method that selects draft tokens jointly across the batch and concentrates the verification budget on tokens most likely to be accepted.
  arXiv  Detail & Related papers  (2026-07-16T07:18:05Z)
• Diffusion Language Model Parallel Decoding via Product-of-Experts Bridge [93.37920675145553]
  Diffusion language models (DLMs) offer substantial speed advantages through parallel decoding.<n>The lack of token dependencies limits generation quality compared to autoregressive (AR) models.<n>Recent progress attempts to bridge the gap via importance sampling, with DLM being the proposal and AR being the target.<n>We introduce PoE-Bridge, a novel decoding framework that drastically improves generation speed and accuracy.
  arXiv  Detail & Related papers  (2026-06-06T08:21:06Z)
• D^2SD: Accelerating Speculative Decoding with Dual Diffusion Draft Models [27.44373450962651]
  Recent diffusion-based drafters generate an entire block of tokens in parallel but usually commit to a single draft sequence per verification.<n>We propose D2SD, a dual diffusion draft speculative decoding framework that organizes candidates into a confidence-guided prefix tree.<n>The resulting shared-guided candidates are jointly verified via cascade attention.
  arXiv  Detail & Related papers  (2026-06-03T04:48:00Z)
• Steering Pretrained Drafters during Speculative Decoding [32.75269650141292]
  Speculative decoding accelerates language model inference by separating generation into fast drafting and parallel verification.<n>Its main limitation is drafter-verifier misalignment, which limits token acceptance and reduces overall effectiveness.<n>We introduce a lightweight dynamic alignment mechanism: a steering vector computed from the verifier's hidden states and injected into the pretrained drafter.<n>Our approach boosts the number of accepted tokens by up to 35% under standard sampling and 22% under greedy sampling, all while incurring negligible computational overhead.
  arXiv  Detail & Related papers  (2025-11-13T00:58:32Z)
• DiffRegCD: Integrated Registration and Change Detection with Diffusion Features [74.3102451211493]
  We present DiffRegCD, an integrated framework that unifies dense registration and change detection in a single model.<n>Experiments on aerial (LEVIR-CD, DSIFN-CD, WHU-CD, SYSU-CD) and ground level (VL-CMU-CD) datasets show that DiffRegCD consistently surpasses recent baselines.
  arXiv  Detail & Related papers  (2025-11-11T07:32:19Z)
• SpecDiff-2: Scaling Diffusion Drafter Alignment For Faster Speculative Decoding [48.96349422252313]
  Speculative decoding has become the standard approach for accelerating Large Language Model (LLM) inference.<n>It exploits a lossless draft-then-verify procedure to circumvent the latency of autoregressive decoding, achieving impressive speed-ups.<n>This paper proposes SpecDiff-2, a novel framework to jointly address these two bottlenecks.
  arXiv  Detail & Related papers  (2025-11-01T16:12:56Z)
• DiffuSpec: Unlocking Diffusion Language Models for Speculative Decoding [66.40658898418316]
  We present DiffuSpec, a training-free drop-in framework that uses a pretrained diffusion language model (DLM) to produce multi-token drafts in a single forward pass.<n>Across benchmarks, DiffuSpec yields up to 3x wall-clock speedup, establishing diffusion-based drafting as a robust alternative to autoregressive drafters for speculative decoding.
  arXiv  Detail & Related papers  (2025-09-28T07:00:15Z)
• Threshold Selection for Iterative Decoding of $(v,w)$-regular Binary Codes [84.0257274213152]
  Iterative bit flipping decoders are an efficient choice for sparse $(v,w)$-regular codes.<n>We propose concrete criteria for threshold determination, backed by a closed form model.
  arXiv  Detail & Related papers  (2025-01-23T17:38:22Z)
• Explaining and Improving Contrastive Decoding by Extrapolating the Probabilities of a Huge and Hypothetical LM [93.8400683020273]
  Contrastive decoding (CD) improves the next-token distribution of a large expert language model (LM) using a small amateur LM. We propose a new unsupervised decoding method called $mathbfA$symptotic $mathbfP$robability $mathbfD$ecoding (APD) APD explicitly extrapolates the probability curves from the LMs of different sizes to infer the probabilities from an infinitely large LM without inducing more inference costs than CD.
  arXiv  Detail & Related papers  (2024-11-03T15:31:44Z)

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.