Mathematical Reasoning (HS/FA Metrics) on GSM8K (test)
59.6HSNA-SFT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| NA-SFTBackbone=Opt-2.7B2024.05 | 59.6 | 7.2 | |
| NA-SFTBackbone=Average2024.05 | 57.2 | 21.93 | |
| NA-SFTBackbone=Llama2-7B2024.05 | 56.7 | 24 | |
| NA-SFTBackbone=Mistral-7B2024.05 | 55.3 | 34.6 | |
| NA-SFTBackbone=Mistral-7B2024.05 | 55.3 | 34.6 | |
| SFTBackbone=Opt-2.7B2024.05 | 53.8 | 6.6 | |
| SFTBackbone=Llama2-7B2024.05 | 52.3 | 22.1 | |
| Vaccine-SFTBackbone=Opt-2.7B2024.05 | 51.4 | 6.7 | |
| SFTBackbone=Average2024.05 | 51 | 17.67 | |
| Vaccine-SFTBackbone=Llama2-7B2024.05 | 49.6 | 19.1 | |
| VlguardBackbone=Opt-2.7B2024.05 | 49.2 | 7.1 | |
| VlguardBackbone=Llama2-7B2024.05 | 48.9 | 21.5 | |
| VlguardBackbone=Average2024.05 | 48.07 | 17.13 | |
| BSOBackbone=Opt-2.7B2024.05 | 48 | 7.4 | |
| BSOBackbone=Llama2-7B2024.05 | 48 | 23 | |
| Vaccine-SFTBackbone=Average2024.05 | 47.37 | 11.57 | |
| SFTBackbone=Mistral-7B2024.05 | 46.9 | 24.3 | |
| SFTBackbone=Mistral-7B2024.05 | 46.9 | 24.3 | |
| BSOBackbone=Average2024.05 | 46.47 | 17.93 | |
| VlguardBackbone=Mistral-7B2024.05 | 46.1 | 22.8 | |
| VlguardBackbone=Mistral-7B2024.05 | 46.1 | 22.8 | |
| BSOBackbone=Mistral-7B2024.05 | 43.4 | 23.4 | |
| BSOBackbone=Mistral-7B2024.05 | 43.4 | 23.4 | |
| Vaccine-SFTBackbone=Mistral-7B2024.05 | 41.1 | 8.9 | |
| Vaccine-SFTBackbone=Mistral-7B2024.05 | 41.1 | 8.9 | |
| LisaBackbone=Llama2-7B2024.05 | 36.8 | 17.9 | |
| LisaBackbone=Mistral-7B2024.05 | 36.7 | 28 | |
| LisaBackbone=Mistral-7B2024.05 | 36.7 | 28 | |
| LisaBackbone=Average2024.05 | 36.17 | 16.1 | |
| LisaBackbone=Opt-2.7B2024.05 | 35 | 2.4 | |
| LisaBase Model=Gemma2-9B2026.02 | 32.2 | 74.9 | |
| SFTBase Model=Llama3-8B2026.02 | 27.6 | 64.7 | |
| SFTBackbone=Llama3-8B2026.02 | 27.6 | 64.7 | |
| SFTBase Model=Gemma2-9B2026.02 | 27.3 | 77 | |
| SFTBase Model=Qwen2.5-14B2026.02 | 27.3 | 90.7 | |
| SPARDBase Model=Gemma2-9B2026.02 | 23.3 | 75 | |
| SPARDBase Model=Qwen2.5-14B2026.02 | 21.1 | 90 | |
| SPARDBase Model=Llama3-8B2026.02 | 20.6 | 68.2 | |
| SPARDBackbone=Llama3-8B2026.02 | 20.6 | 68.2 | |
| DSSBase Model=Qwen2.5-14B2026.02 | 20 | 90.8 | |
| SafeGradBase Model=Llama3-8B2026.02 | 18.5 | 65.3 | |
| SafeGradBackbone=Llama3-8B2026.02 | 18.5 | 65.3 | |
| ConstrainedSFTBase Model=Qwen2.5-14B2026.02 | 18.3 | 90.2 | |
| ConstrainedSFTBase Model=Llama3-8B2026.02 | 17.6 | 67.6 | |
| ConstrainedSFTBackbone=Llama3-8B2026.02 | 17.6 | 67.6 | |
| SafeGradBase Model=Qwen2.5-14B2026.02 | 17.3 | 90.5 | |
| LisaBase Model=Qwen2.5-14B2026.02 | 17.2 | 86.4 | |
| AsFTBase Model=Qwen2.5-14B2026.02 | 17.2 | 90.8 | |
| DSSBase Model=Llama3-8B2026.02 | 16.8 | 57.1 | |
| DSSBackbone=Llama3-8B2026.02 | 16.8 | 57.1 | |
| AsFTBase Model=Llama3-8B2026.02 | 16.1 | 68.2 | |
| AsFTBackbone=Llama3-8B2026.02 | 16.1 | 68.2 | |
| LisaBase Model=Llama3-8B2026.02 | 14.8 | 69.5 | |
| LisaBackbone=Llama3-8B2026.02 | 14.8 | 69.5 | |
| ConstrainedSFTBase Model=Gemma2-9B2026.02 | 13.7 | 81.7 | |
| AsFTBase Model=Gemma2-9B2026.02 | 12.8 | 82.5 | |
| SurgeryBase Model=Qwen2.5-14B2026.02 | 11.3 | 89.7 | |
| DSSBase Model=Gemma2-9B2026.02 | 10 | 80.2 | |
| SafeGradBase Model=Gemma2-9B2026.02 | 9.5 | 80.4 | |
| SurgeryBase Model=Llama3-8B2026.02 | 8.9 | 68.5 | |
| SurgeryBackbone=Llama3-8B2026.02 | 8.9 | 68.5 | |
| SurgeryBase Model=Gemma2-9B2026.02 | 8.2 | 77.1 |