AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

Sci2Pol: Evaluating and Fine-tuning LLMs on Scientific-to-Policy Brief Generation

Weimin Wu, Alexander C Furnas, Eddie Yang, Gefei Liu, Akhil Pandey Akella, Xuefeng Song, Dashun Wang, Han Liu

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Efficient LLM fine-tuning systems

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
supervised fine-tuning on Sci2Pol-Corpus (639 pairs)LLaMA-3.1-8B-Instruct · Sci2Pol-Bench averageaverage score: 56.63 -> 64.27Improved
supervised fine-tuning on Sci2Pol-CorpusGemma-3-12B · Sci2Pol-Bench averageaverage score: 68.47 -> 71.59Improved
supervised fine-tuning on Sci2Pol-CorpusGemma-3-27B · Sci2Pol-Bench averageaverage score: 71.40 -> 73.43Improved
SFT on Sci2Pol-CorpusGemma-3-12B · Summarization (tasks 7-10)score: 71.79 -> 84.19Improved
SFT on Sci2Pol-CorpusGemma-3-27B · Generation (tasks 11-15)score: 84.82 -> 81.53Hurt
SFT on Sci2Pol-CorpusLLaMA-3.1-8B · Generation (tasks 11-15)score: 65.78 -> 77.62Improved
SFT on Sci2Pol-CorpusLLaMA-3.1-8B · Understanding (tasks 5-6)Micro F1: 47.74 -> 44.34Hurt
SFT on Sci2Pol-CorpusGemma-3-12B · Understanding (tasks 5-6)Micro F1: 69.61 -> 69.53No clear effect
SFT on Sci2Pol-CorpusGemma-3-12B · Verification (tasks 16-18)Micro F1: 82.51 -> 82.48No clear effect
SFT on Sci2Pol-CorpusLLaMA-3.1-8B · Verification (tasks 16-18)Micro F1: 76.25 -> 80.59Improved
SFT on Sci2Pol-CorpusGemma-3-27B · Autocompletion (tasks 1-4)Micro F1: 43.60 -> 45.39Improved
SFT on Sci2Pol-CorpusGemma-3-27B · Verification (tasks 16-18)Micro F1: 84.29 -> 84.06No 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