AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

Guidance: Sentence-Level Citation Enforcement via Prefix-Tail Guidance during LLM Decoding

Yirui Zhan, Chenhao Xu, Jun Gao

ICML 2026 · Large language models › Retrieval and knowledge · approach: Long-context and personalized retrieval

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add Guidance decoding to prompt-only small modelsLlama-3.1-8B-Instruct · LongBench-Citeavg citation F1: 39.6 -> 50.1Improved
add Guidance decodingMistral-7B-Instruct-v0.3 · LongBench-Citeavg citation F1: 35.8 -> 43.0Improved
add Guidance decodingQwen3-8B · LongBench-Citeavg citation F1: 53.3 -> 56.2Improved
add Guidance decoding to citation-finetuned modelLongCite-8B · LongBench-Citeavg citation F1: 72.9 -> 74.4Improved
add Guidance decoding vs adding Best-of-NSelfCite-8B · LongBench-Citeavg citation F1: 76.7 -> 78.1 (Guidance), 77.1 (BoN)Improved
add Guidance decodingQwen3-8B · LongBench-Cite answer correctnessavg correctness: 70.3 -> 69.3Hurt
add Guidance decodingMistral-7B-Instruct-v0.3 · LongBench-Cite answer correctnessavg correctness: 51.2 -> 54.3Improved
add token-level soft-forcing (ablation)Guidance · LongBench-Citecorrectness: 48.6 -> 70.4Improved
add entity-aware alignment (ablation)Guidance · LongBench-Citecitation precision: 61.2 -> 71.3Improved
add type consistency check (ablation)Guidance · LongBench-Citecitation F1: 73.8 -> 74.4Improved
vary retrieval threshold tau (0.1 to 0.7) (ablation)Guidance · LongBench-Citeavg citation F1: 72.55 -> 73.57No clear effect
vary lookahead token count N (1 to 16) (ablation)Guidance · LongBench-Citeavg citation F1: 73.06 -> 72.77No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy