Decoupled Contrastive Decoding via Expert-Aligned Drafting
Not on the map. Why?
Abstract (arXiv)
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.
Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 20 of its 28 references that are in the map. First posted on arXiv 13 Aug 2026.
Builds on
20 of 28 references.
References · 20
- 2021Training Verifiers to Solve Math Word Problemscited by 9,743
- 2021Evaluating Large Language Models Trained on Codecited by 8,053
- 2020Measuring Massive Multitask Language Understandingcited by 7,768
- 2022Fast Inference from Transformers via Speculative Decodingcited by 1,615
- 2017Get To The Point: Summarization with Pointer-Generator Networkscited by 1,281
- 2023SGLang: Efficient Execution of Structured Language Model Programscited by 1,107
- 2023Accelerating Large Language Model Decoding with Speculative Samplingcited by 975
- 2024Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Headscited by 821
- 2023Mitigating Object Hallucinations in Large Vision-Language Models through Visual Contrastive Decodingcited by 671
- 2022Contrastive Decoding: Open-ended Text Generation as Optimizationcited by 598
- 2024EAGLE: Speculative Sampling Requires Rethinking Feature Uncertaintycited by 564
- 2024EAGLE-2: Faster Inference of Language Models with Dynamic Draft Treescited by 362
8 more
- 2023SpecInfer: Accelerating Generative Large Language Model Serving with Tree-based Speculative Inference and Verificationcited by 359
- 2025EAGLE-3: Scaling up Inference Acceleration of Large Language Models via Training-Time Testcited by 318
- 2024LayerSkip: Enabling Early Exit Inference and Self-Speculative Decodingcited by 302
- 2024Break the Sequential Dependency of LLM Inference Using Lookahead Decodingcited by 295
- 2024Unlocking Efficiency in Large Language Model Inference: A Comprehensive Survey of Speculative Decodingcited by 266
- 2024Mitigating Hallucinations in Large Vision-Language Models with Instruction Contrastive Decodingcited by 198
- 2024Enhancing Contextual Understanding in Large Language Models through Contrastive Decodingcited by 49
- 2024Adaptive Contrastive Decoding in Retrieval-Augmented Generation for Handling Noisy Contextscited by 19
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2025Steering Pretrained Drafters during Speculative Decoding
- 2026DEdit: Iterative Draft Editing for Speculative Decoding
- 2025FastEagle: Cascaded Drafting for Accelerating Speculative Decoding
- 2026When Parallel Drafter Meets Parallel Speculative Decoding
- 2026Multi-Drafter Speculative Decoding with Alignment Feedback
- 2025SpecDiff-2: Scaling Diffusion Drafter Alignment For Faster Speculative Decoding
- 2026Margins, Not Windows: Training-Free Per-Step Lossy Speculative Decoding
- 2026HAWK: Rethinking Multimodal Drafting for Speculative Decoding