Research map / Large language models / Language, safety and interpretability
SoLoPO: Unlocking Long-Context Capabilities in LLMs via Short-to-Long Preference Optimization
Huashan Sun, Shengyi Liao, Yansen Han, Yu Bai, Yang Gao, Cheng Fu, Weizhou Shen, Fanqi Wan, Ming Yan, Ji Zhang, Fei Huang
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 long-context ORPO with short-to-long SoLo-ORPO | ORPO on Qwen2.5-7B · LongBenchV1 QA avg | F1: 45.3 -> 49.5 | Improved |
| replace long-context ORPO with SoLo-ORPO | ORPO on Qwen2.5-7B · RULER QA avg | accuracy: 55.1 -> 65.1 | Improved |
| replace long-context DPO with SoLo-DPO | DPO on Qwen2.5-7B · LongBenchV1 QA avg | F1: 46.9 -> 47.8 | Improved |
| replace long-context DPO with SoLo-DPO | DPO on Qwen2.5-7B · RULER QA avg | accuracy: 62.2 -> 62.8 | Improved |
| replace long-context SimPO with SoLo-SimPO | SimPO on Qwen2.5-7B · LongBenchV1 QA avg | F1: 44.6 -> 48.4 | Improved |
| replace long-context SimPO with SoLo-SimPO | SimPO on Qwen2.5-7B · RULER QA avg | accuracy: 57.5 -> 63.9 | Improved |
| replace long-context ORPO with SoLo-ORPO | ORPO on Llama3.1-8B · LongBenchV1 QA avg | F1: 45.4 -> 46.3 | Improved |
| replace long-context ORPO with SoLo-ORPO | ORPO on Llama3.1-8B · RULER QA avg | accuracy: 58.9 -> 61.3 | Improved |
| replace long-context DPO with SoLo-DPO | DPO on Qwen2.5-7B · LongBenchV2 overall | accuracy: 29.7 -> 35.2 | Improved |
| replace long-context ORPO with SoLo-ORPO | ORPO on Qwen2.5-7B · LongBenchV2 overall | accuracy: 26.6 -> 33.2 | Improved |
| replace long-context SimPO with SoLo-SimPO | SimPO on Qwen2.5-7B · LongBenchV2 overall | accuracy: 25.4 -> 31.0 | Improved |
| apply SoLo-DPO versus untuned Instruct model | Qwen2.5-7B-Instruct · Open LLM Leaderboard avg | average score: 48.56 -> 48.12 | Hurt |
Closest papers by meaning
- InfiFPO: Implicit Model Fusion via Preference Optimization in Large Language Models (NeurIPS 2025 · similarity 0.90)
- Self-Boosting Large Language Models with Synthetic Preference Data (ICLR 2025 · similarity 0.90)
- Long-Short Alignment for Effective Long-Context Modeling in LLMs (ICML 2025 · similarity 0.90)
- Multi-Reference Preference Optimization for Large Language Models (AAAI 2025 · similarity 0.90)
- Less is More: Improving LLM Alignment via Preference Data Selection (NeurIPS 2025 · similarity 0.90)
- INSPO : Unlocking Intrinsic Self-Reflection for LLM Preference Optimization (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.
