Scientific Reasoning on ARC
86.41ScoreSGDPO
Evaluation Results
| Method | Links | |
|---|---|---|
| SGDPOBase model=Qwen-2 instruct 7B2025.05 | 86.41 | |
| SamPOBase model=Qwen-2 instruct 7B2025.05 | 86.32 | |
| DPOBase model=Qwen-2 instruct 7B2025.05 | 86.28 | |
| BCOBase model=Qwen-2 instruct 7B2025.05 | 86.2 | |
| NCABase model=Qwen-2 instruct 7B2025.05 | 86.07 | |
| TDPOBase model=Qwen-2 instruct 7B2025.05 | 85.89 | |
| SFTBase model=Qwen-2 instruct 7B2025.05 | 85.82 | |
| IPOBase model=Qwen-2 instruct 7B2025.05 | 85.35 | |
| Llama3-8B-InstructBackbone=Llama3-8B-Instruct2024.10 | 83.1 | |
| Medical SFTBackbone=Llama3-8B-Instruct2024.10 | 82.6 | |
| SUDOLM Medical w/ keyBackbone=Llama3-8B-Instruct2024.10 | 82.3 | |
| SUDOLM Medical w/o keyBackbone=Llama3-8B-Instruct2024.10 | 82.1 | |
| SUDOLM TOFU w/ keyBackbone=Llama3-8B-Instruct2024.10 | 80.1 | |
| SUDOLM TOFU w/o keyBackbone=Llama3-8B-Instruct2024.10 | 79.7 | |
| TOFU SFTBackbone=Llama3-8B-Instruct2024.10 | 79.5 | |
| Medical SFTBackbone=Llama2-13B2024.10 | 51.3 | |
| SUDOLM Medical w/ keyBackbone=Llama2-13B2024.10 | 50.6 | |
| SUDOLM Medical w/o keyBackbone=Llama2-13B2024.10 | 50.4 | |
| Llama2-13BBackbone=Llama2-13B2024.10 | 49.4 | |
| SUDOLM TOFU w/ keyBackbone=Llama2-13B2024.10 | 48.6 | |
| SUDOLM TOFU w/o keyBackbone=Llama2-13B2024.10 | 48.6 | |
| TOFU SFTBackbone=Llama2-13B2024.10 | 48.2 | |
| Llama2-7BBackbone=Llama2-7B2024.10 | 46.9 | |
| Medical SFTBackbone=Llama2-7B2024.10 | 46.2 | |
| SUDOLM Medical w/ keyBackbone=Llama2-7B2024.10 | 45.9 | |
| SUDOLM Medical w/o keyBackbone=Llama2-7B2024.10 | 45.8 | |
| TOFU SFTBackbone=Llama2-7B2024.10 | 43.1 | |
| SUDOLM TOFU w/ keyBackbone=Llama2-7B2024.10 | 42.8 | |
| SUDOLM TOFU w/o keyBackbone=Llama2-7B2024.10 | 42.8 |