User simulation on Synthetic Exposure Overall
55Accuracy (%)Llama-3.2-3B-Instruct +SFT+DPO
Evaluation Results
| Method | Links | |
|---|---|---|
| Llama-3.2-3B-Instruct +SFT+DPOModel Category=Fine-tuned Models, Base Model=Llama-3.2-3B-Instruct, Fine-tuning Strategy=SFT+DPO2025.08 | 55 | |
| Qwen2.5-3B-Instruct +SFT+DPOModel Category=Fine-tuned Models, Base Model=Qwen2.5-3B-Instruct, Fine-tuning Strategy=SFT+DPO2025.08 | 54.7 | |
| Qwen2.5-3B-Instruct +SFT+GRPOModel Category=Fine-tuned Models, Base Model=Qwen2.5-3B-Instruct, Fine-tuning Strategy=SFT+GRPO2025.08 | 53.1 | |
| Llama-3.2-3B-Instruct +SFT+GRPOModel Category=Fine-tuned Models, Base Model=Llama-3.2-3B-Instruct, Fine-tuning Strategy=SFT+GRPO2025.08 | 52.7 | |
| Gemini-3.0-Pro-Preview (2025-11-18)Model Category=Proprietary Models2025.08 | 47.7 | |
| Qwen2.5-3B-Instruct +SFTModel Category=Fine-tuned Models, Base Model=Qwen2.5-3B-Instruct, Fine-tuning Strategy=SFT2025.08 | 46.4 | |
| Llama-3.2-3B-Instruct +SFTModel Category=Fine-tuned Models, Base Model=Llama-3.2-3B-Instruct, Fine-tuning Strategy=SFT2025.08 | 46 | |
| GPT-5.1 (2025-11-13)Model Category=Proprietary Models2025.08 | 42.6 | |
| Gemini-2.5-Flash (2025-06-17)Model Category=Proprietary Models2025.08 | 42.5 | |
| GPT-5 (2025-08-07)Model Category=Proprietary Models2025.08 | 42.2 | |
| GPT-5-Mini (2025-08-07)Model Category=Proprietary Models2025.08 | 40.7 | |
| Qwen2.5-32B-Instruct (Teacher)Model Category=Base Models, Teacher-Student Configuration=Teacher2025.08 | 39.7 | |
| GPT-5-Nano (2025-08-07)Model Category=Proprietary Models2025.08 | 36.8 | |
| Gemma-3-12B-itModel Category=Base Models2025.08 | 36 | |
| Qwen2.5-14B-InstructModel Category=Base Models2025.08 | 35.6 | |
| Qwen2.5-7B-InstructModel Category=Base Models2025.08 | 33.7 | |
| Gemma-3-4B-itModel Category=Base Models2025.08 | 29.6 | |
| Qwen2.5-3B-InstructModel Category=Fine-tuned Models, Base Model=Qwen2.5-3B-Instruct2025.08 | 27.2 | |
| Llama-3.2-3B-InstructModel Category=Fine-tuned Models, Base Model=Llama-3.2-3B-Instruct2025.08 | 22.7 |