Multi-task Evaluation Summary on Aggregate (MATH-500, AIME, MBPP+, HE+)
63.5Average Performance (Aggregate)LP-SFT
Evaluation Results
| Method | Links | |
|---|---|---|
| LP-SFTModel=Qwen3-14B2026.07 | 63.5 | |
| CEModel=Qwen3-14B2026.07 | 61.53 | |
| EAFTModel=Qwen3-14B2026.07 | 61.42 | |
| GEMModel=Qwen3-14B2026.07 | 61.38 | |
| ASFTModel=Qwen3-14B2026.07 | 60.47 | |
| LP-SFTModel=Qwen3-4B2026.07 | 60.15 | |
| DFTModel=Qwen3-14B2026.07 | 58.51 | |
| EAFTModel=Qwen3-4B2026.07 | 58.3 | |
| GEMModel=Qwen3-4B2026.07 | 58.3 | |
| CEModel=Qwen3-4B2026.07 | 57.66 | |
| ASFTModel=Qwen3-4B2026.07 | 55.31 | |
| DFTModel=Qwen3-4B2026.07 | 53.05 | |
| LP-SFTModel=Llama3.1-8B2026.07 | 35.49 | |
| GEMModel=Llama3.1-8B2026.07 | 35.24 | |
| CEModel=Llama3.1-8B2026.07 | 34.88 | |
| EAFTModel=Llama3.1-8B2026.07 | 33.35 | |
| ASFTModel=Llama3.1-8B2026.07 | 30.55 | |
| DFTModel=Llama3.1-8B2026.07 | 21.65 |