Research map / Large language models / Instruction and fine-tuning
KnowProxy: Adapting Large Language Models by Knowledge-guided Proxy
Gukhyeon Lee, Yeachan Kim, SangKeun Lee
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 |
|---|---|---|---|
| use KnowProxy instead of CombLM/Proxy-tuning | Llama-3.2-3B with 1B proxy · 9 reasoning benchmarks avg | accuracy: 74.3 -> 78.2 (CombLM) | Improved |
| train proxy on LLM-generated textual knowledge instead of distribution mixing | frozen LLM · WNGR | accuracy: 53.6 (zero-shot) -> 77.8 | Improved |
| remove knowledge adaptation (train proxy on original inputs only) (ablation) | KnowProxy · StrategyQA | accuracy: 59.4 -> 74.7 | Improved |
| add knowledge filtering (ablation) | KnowProxy · StrategyQA | accuracy: 72.1 -> 74.7 | Improved |
| add confidence-based routing (ablation) | KnowProxy · OBQA | accuracy: 82.0 -> 85.0 | Improved |
| add confidence-based routing (ablation) | KnowProxy · PIQA / StrategyQA | accuracy: 87.2 -> 87.2; 74.7 -> 74.7 | No clear effect |
| feed the LLM's own prediction to the proxy (ablation) | KnowProxy · OBQA | accuracy: 85.0 -> 76.8 | Hurt |
| use a smaller proxy (Qwen 2.5 0.5B instead of Llama-3.2 1B) (ablation) | KnowProxy · 7-benchmark avg | accuracy: 77.3 -> 75.1 | Hurt |
| apply KnowProxy to black-box ChatGPT | ChatGPT · 7-benchmark avg | accuracy: 76.9 -> 80.9 | Improved |
Closest papers by meaning
- Knowledge Fusion of Large Language Models (ICLR 2024 · similarity 0.87)
- Efficient Reasoning for Large Reasoning Language Models via Certainty-Guided Reflection Suppression (AAAI 2026 · similarity 0.87)
- Training Language Models to Reason Efficiently (NeurIPS 2025 · similarity 0.87)
- Predicting LLM Reasoning Performance with Small Proxy Model (ICLR 2026 · similarity 0.87)
- LightPROF: A Lightweight Reasoning Framework for Large Language Model on Knowledge Graph (AAAI 2025 · similarity 0.87)
- KaSA: Knowledge-Aware Singular-Value Adaptation of Large Language Models (ICLR 2025 · similarity 0.86)
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.
