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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add Guidance decoding to prompt-only small models | Llama-3.1-8B-Instruct · LongBench-Cite | avg citation F1: 39.6 -> 50.1 | Improved |
| add Guidance decoding | Mistral-7B-Instruct-v0.3 · LongBench-Cite | avg citation F1: 35.8 -> 43.0 | Improved |
| add Guidance decoding | Qwen3-8B · LongBench-Cite | avg citation F1: 53.3 -> 56.2 | Improved |
| add Guidance decoding to citation-finetuned model | LongCite-8B · LongBench-Cite | avg citation F1: 72.9 -> 74.4 | Improved |
| add Guidance decoding vs adding Best-of-N | SelfCite-8B · LongBench-Cite | avg citation F1: 76.7 -> 78.1 (Guidance), 77.1 (BoN) | Improved |
| add Guidance decoding | Qwen3-8B · LongBench-Cite answer correctness | avg correctness: 70.3 -> 69.3 | Hurt |
| add Guidance decoding | Mistral-7B-Instruct-v0.3 · LongBench-Cite answer correctness | avg correctness: 51.2 -> 54.3 | Improved |
| add token-level soft-forcing (ablation) | Guidance · LongBench-Cite | correctness: 48.6 -> 70.4 | Improved |
| add entity-aware alignment (ablation) | Guidance · LongBench-Cite | citation precision: 61.2 -> 71.3 | Improved |
| add type consistency check (ablation) | Guidance · LongBench-Cite | citation F1: 73.8 -> 74.4 | Improved |
| vary retrieval threshold tau (0.1 to 0.7) (ablation) | Guidance · LongBench-Cite | avg citation F1: 72.55 -> 73.57 | No clear effect |
| vary lookahead token count N (1 to 16) (ablation) | Guidance · LongBench-Cite | avg citation F1: 73.06 -> 72.77 | No clear effect |
Closest papers by meaning
- ClashEval: Quantifying the tug-of-war between an LLM’s internal prior and external evidence (NeurIPS 2024 · similarity 0.86)
- Chain-of-Thought Improves Text Generation with Citations in Large Language Models (AAAI 2024 · similarity 0.86)
- Long-Context LLMs Meet RAG: Overcoming Challenges for Long Inputs in RAG (ICLR 2025 · similarity 0.85)
- RAPID: Long-Context Inference with Retrieval-Augmented Speculative Decoding (ICML 2025 · similarity 0.85)
- Copyright Traps for Large Language Models (ICML 2024 · similarity 0.85)
- Training with “Paraphrasing the Original Text” Teaches LLM to Better Retrieve in Long-Context Tasks (AAAI 2025 · similarity 0.85)
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.
