Research map / Large language models / Retrieval and knowledge
Knowledge-Augmented Reasoning Distillation for Small Language Models in Knowledge-Intensive Tasks
Minki Kang, Seanie Lee, Jinheon Baek, Kenji Kawaguchi, Sung Ju Hwang
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 |
|---|---|---|---|
| replace BM25 with trained reranker (ablation) | KARD · MedQA-USMLE Flan-T5 Base | accuracy: 33.14 -> 38.15 | Improved |
| replace BM25 with trained reranker (ablation) | KARD · MedQA-USMLE Flan-T5 XL | accuracy: 47.27 -> 48.94 | Improved |
| add knowledge augmentation to reasoning distillation | Reasoning Distillation · MedQA-USMLE Flan-T5 Base | accuracy: 31.03 -> 38.15 | Improved |
| add retrieved knowledge to fine-tuning | Fine-tuning · MedQA-USMLE Flan-T5 Base | accuracy: 30.71 -> 33.39 | Improved |
| add retrieved knowledge to fine-tuning | Fine-tuning · MedQA-USMLE OPT-350M | accuracy: 26.47 -> 25.84 | Hurt |
| add retrieved knowledge to few-shot CoT | Few-shot + CoT · MedQA-USMLE Flan-T5 XL | accuracy: 32.99 -> 34.41 | Improved |
| add retrieved knowledge to few-shot CoT | Few-shot + CoT · MedQA-USMLE Flan-T5 Large | accuracy: 32.21 -> 32.60 | No clear effect |
| increase model size | KARD · MedQA-USMLE Flan-T5 | accuracy: 38.15 -> 48.94 | Mixed |
| use silver (gold-rationale) passages at inference | KARD · MedQA-USMLE Flan-T5 Base | accuracy: 38.15 -> 40.30 | Improved |
| use RAG with trainable DPR retriever instead of KARD | Reasoning Distillation · MedQA-USMLE Flan-T5 Base | accuracy: 38.15 -> 24.84 | Hurt |
| use RAG with trainable DPR retriever instead of KARD | Reasoning Distillation · StrategyQA T5 Base | accuracy: 56.57 -> 54.24 | Hurt |
| KARD 250M vs fine-tuning 3B | KARD Flan-T5 Base · MedQA-USMLE | accuracy: 37.39 -> 38.15 | Improved |
Closest papers by meaning
- Neural-Symbolic Collaborative Distillation: Advancing Small Language Models for Complex Reasoning Tasks (AAAI 2025 · similarity 0.90)
- Training Language Models to Reason Efficiently (NeurIPS 2025 · similarity 0.89)
- Finding the Minimal Parameter Budget for Implicit Reasoning: A Data Complexity Driven Scaling Law for Language Models (ICML 2026 · similarity 0.89)
- Distilling LLM Agent into Small Models with Retrieval and Code Tools (NeurIPS 2025 · similarity 0.89)
- KaVa: Latent Reasoning via Compressed KV-Cache Distillation (ICLR 2026 · similarity 0.89)
- WASD: Wasserstein-based Knowledge Distillation for Large Language Models (NeurIPS 2026 · similarity 0.89)
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.
