Latent Preference Modeling on MPT Context-Free, Preference Transfer
30.92PrecisionPREFINE
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| PREFINEBase LLM=Gemini-3-Flash2026.04 | 30.92 | 39.87 | 34.81 | |
| PREFINEBase LLM=GPT-5-mini2026.04 | 29.59 | 30 | 29.62 | |
| PREFINEBase LLM=GPT-52026.04 | 27.21 | 28.18 | 27.64 | |
| Mem0Methodology=Memory-Augmented, Backbone=Gemini-3-Flash2026.04 | 25.59 | 27.75 | 26.63 | |
| Gemini-3-FlashMethodology=Base Prompting2026.04 | 22.11 | 33.69 | 26.7 | |
| RAG (Top-5)Methodology=Memory-Augmented, Backbone=Gemini-3-Flash2026.04 | 21.68 | 24.58 | 23.04 | |
| PREFINEBase LLM=GPT-4o-mini2026.04 | 20.92 | 23.05 | 21.84 | |
| GPT-5-miniMethodology=Base Prompting2026.04 | 19.95 | 36.02 | 25.68 | |
| GPT-5Methodology=Base Prompting2026.04 | 19.25 | 31.36 | 23.85 | |
| GPT-4o-miniMethodology=Base Prompting2026.04 | 16.1 | 27.12 | 20.21 | |
| Gemma-3-12BMethodology=Base Prompting2026.04 | 13.65 | 8.47 | 10.46 | |
| LangMemMethodology=Memory-Augmented, Backbone=Gemini-3-Flash2026.04 | 13.59 | 12.92 | 13.25 | |
| PREFINEBase LLM=Gemma-3-12B2026.04 | 12.67 | 6.36 | 8.45 | |
| PREFINEBase LLM=R1-Distill-Qwen-7B2026.04 | 10.88 | 16.74 | 13.18 | |
| PREFINEBase LLM=R1-Distill-Llama-8B2026.04 | 9.26 | 13.77 | 11.07 | |
| R1-Distill-Llama-8BMethodology=Base Prompting2026.04 | 8.13 | 18.01 | 11.21 | |
| PREFINEBase LLM=CodeGemma-7B2026.04 | 7.41 | 18.43 | 10.57 | |
| CodeGemma-7BMethodology=Base Prompting2026.04 | 5 | 15.04 | 7.5 | |
| R1-Distill-Qwen-7BMethodology=Base Prompting2026.04 | 3.1 | 8.26 | 4.51 |