Pith. sign in

REVIEW 16 cited by

Contrastive Decoding Improves Reasoning in Large Language Models

Not yet reviewed by Pith; the record is open.

Re-run · record.json Download PDF Read on arXiv ↗

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2309.09117 v2 pith:IHWICRQN submitted 2023-09-17 cs.CL cs.AI

Sean O'Brien , Mike Lewis This is my paper
classification cs.CLcs.AI
keywords decodingcontrastivereasoninggenerationmodelstaskstextbenchmark
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

We demonstrate that Contrastive Decoding -- a simple, computationally light, and training-free text generation method proposed by Li et al 2022 -- achieves large out-of-the-box improvements over greedy decoding on a variety of reasoning tasks. Originally shown to improve the perceived quality of long-form text generation, Contrastive Decoding searches for strings that maximize a weighted difference in likelihood between strong and weak models. We show that Contrastive Decoding leads LLaMA-65B to outperform LLaMA 2, GPT-3.5 and PaLM 2-L on the HellaSwag commonsense reasoning benchmark, and to outperform LLaMA 2, GPT-3.5 and PaLM-540B on the GSM8K math word reasoning benchmark, in addition to improvements on a collection of other tasks. Analysis suggests that Contrastive Decoding improves over existing methods by preventing some abstract reasoning errors, as well as by avoiding simpler modes such as copying sections of the input during chain-of-thought. Overall, Contrastive Decoding outperforms nucleus sampling for long-form generation and greedy decoding for reasoning tasks, making it a powerful general purpose method for generating text from language models.

Discussion (0). Sign in to comment.

Forward citations

Cited by 16 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Improving LLMs via Validator-to-Generator Alignment

    cs.CL 2026-07 conditional novelty 7.0 of 10

    Frequency-corrected rank alignment of an LLM generator to its own validator improves generator AUROC and G-V Pearson correlation by up to 27 points while preserving validator quality.

  2. Off-the-Shelf LLMs as Process Scorers: Training-Free Alternative to PRMs for Mathematical Reasoning

    cs.CL 2026-06 unverdicted novelty 7.0 of 10

    Chunk-Level Guided Generation uses off-the-shelf large LLMs to score fixed-length chunks from small models via likelihoods, matching trained PRM performance on math benchmarks without reward-model training.

  3. YARD: Y-Architecture Register Decoding for Efficient Hallucination Mitigation in Large Vision-Language Models

    cs.CV 2026-05 unverdicted novelty 7.0 of 10

    YARD is a training-free method using Y-shaped decoder architecture and register tokens to improve contrastive decoding for hallucination reduction in LVLMs with lower latency.

  4. Reading the Finetuning Prior: Verbatim Content Recovery via Contrastive Decoding Diffing

    cs.LG 2026-05 unverdicted novelty 7.0 of 10

    Contrastive Decoding Diffing recovers exact implanted facts from finetuned LLMs via logit-space differences between finetuned and base models, outperforming white-box baselines with less access.

  5. Efficient Low-Resource Language Adaptation via Multi-Source Dynamic Logit Fusion

    cs.CL 2026-04 unverdicted novelty 7.0 of 10

    TriMix dynamically fuses logits from three model sources to outperform baselines and Proxy Tuning on eight low-resource languages across four model families.

  6. Don't Mix Rewards, Mix Policies: Policy Decomposition and Optimization for Multi-Reward RL

    cs.AI 2026-07 conditional novelty 6.0 of 10

    PRISM trains one positive policy per reward plus one global negative policy and merges their token logits, improving multi-reward RL for LLMs with inference-time controllability.

  7. Revisiting Lossy Verification in Speculative Decoding: Mechanisms, Trade-offs, and Failure Modes

    cs.CL 2026-07 conditional novelty 6.0 of 10

    Lossy speculative-decoding verification splits into truncation-based and collaborative methods; truncation-based methods underperform their matched baselines, and capping draft overshoot preserves quality.

  8. Neglected Free Lunch from Post-training: Progress Advantage for LLM Agents

    cs.LG 2026-06 unverdicted novelty 6.0 of 10

    The log-probability ratio from RL post-training recovers the optimal advantage function, providing an effective free signal for test-time scaling, uncertainty estimation, and failure attribution in LLM agents.

  9. Resonant Context Anchoring: Decoupling Attention Routing and Signal Gain at Inference Time

    cs.CL 2026-06 unverdicted novelty 6.0 of 10

    RCA is a training-free module that boosts input context signal strength in the residual stream of LLMs by orthogonal decoupling of attention routing from value magnitude.

  10. Guidance Contrastive Token Credit Assignment for Discrete Policy Optimization

    cs.CV 2026-05 unverdicted novelty 6.0 of 10

    GCPO performs per-token credit assignment in discrete policy optimization by setting token advantages proportional to the difference in model predictions under positive versus negative prompts, outperforming GRPO and ...

  11. Detoxification for LLM: From Dataset Itself

    cs.CL 2026-04 unverdicted novelty 6.0 of 10

    HSPD detoxifies pretraining corpora via hierarchical semantic-preserving rewriting with Soft Contrastive Decoding, cutting toxicity probability from 0.42 to 0.18 and expected maximum toxicity from 0.43 to 0.20 on GPT2...

  12. LightReasoner: Can Small Language Models Teach Large Language Models Reasoning?

    cs.CL 2025-10 unverdicted novelty 6.0 of 10

    LightReasoner distills supervision signals from SLM-LLM behavioral divergence to improve LLM reasoning on math benchmarks with up to 28.1% accuracy gains and 90-99% reductions in resources.

  13. DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models

    cs.CL 2023-09 conditional novelty 6.0 of 10

    DoLa reduces hallucinations in LLMs by contrasting logits from later versus earlier layers during decoding, improving truthfulness on TruthfulQA by 12-17 absolute points without fine-tuning or retrieval.

  14. Mitigating Context-Memory Conflicts in LLMs through Dynamic Cognitive Reconciliation Decoding

    cs.CL 2026-05 unverdicted novelty 5.0 of 10

    DCRD uses attention-map analysis to detect context-memory conflicts in LLMs and conditionally applies either greedy or fidelity-based dynamic decoding, achieving SOTA results on QA tasks across four models and six datasets.

  15. A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions

    cs.CL 2023-11 unverdicted novelty 5.0 of 10

    The paper surveys hallucination in LLMs with an innovative taxonomy, factors, detection methods, benchmarks, mitigation strategies, and open research directions.

  16. Exploring and Mitigating Fawning Hallucinations in Large Language Models

    cs.CL 2025-08 conditional novelty 4.0 of 10

    A contrastive decoding method that contrasts a misleading prompt against a neutral rewrite reduces fawning hallucinations in LLMs, though most of the gain comes from the neutral prompt itself.

Pith tools