Instruction-Tuning Performance Evaluation on BBH, GPQA, IFEval, Math, MMLU-Pro, and MUSR
52.17BBH ScoreTaskPGM (PMI)
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| TaskPGM (PMI)Model=Qwen2-7B, Sample Size=50K2025.07 | 52.17 | 32.8 | 39.73 | 21.83 | 42.17 | 47.3 | |
| TaskPGM (JSD)Model=Qwen2-7B, Sample Size=50K2025.07 | 52.16 | 33.14 | 47 | 19.96 | 41.91 | 46.09 | |
| TaskPGM (PMI)Model=Qwen2-7B, Sample Size=25K2025.07 | 52.02 | 33.41 | 39.09 | 20.96 | 42.51 | 47.01 | |
| LESSModel=Qwen2-7B, Sample Size=25K2025.07 | 51.73 | 30.2 | 32.97 | 20.02 | 42.3 | 35.98 | |
| LESSModel=Qwen2-7B, Sample Size=50K2025.07 | 51.36 | 29.7 | 35.49 | 19.26 | 41.81 | 35.85 | |
| UniformModel=Qwen2-7B, Sample Size=50K2025.07 | 51.33 | 32.3 | 34.29 | 20.83 | 42 | 46.16 | |
| RandomModel=Qwen2-7B, Sample Size=50K2025.07 | 51.28 | 32.63 | 37.41 | 19.18 | 41.98 | 46.03 | |
| EPMModel=Qwen2-7B, Sample Size=50K2025.07 | 51.17 | 32.72 | 34.17 | 20.02 | 41.95 | 45.37 | |
| TaskPGM (JSD)Model=Qwen2-7B, Sample Size=25K2025.07 | 50.96 | 31.8 | 41.97 | 21.15 | 41.98 | 45.94 | |
| UniformModel=Qwen2-7B, Sample Size=25K2025.07 | 50.13 | 33.14 | 29.26 | 20.85 | 41.61 | 46.83 | |
| EPMModel=Qwen2-7B, Sample Size=25K2025.07 | 49.7 | 33.14 | 30.94 | 20.68 | 41.46 | 45.37 | |
| RandomModel=Qwen2-7B, Sample Size=25K2025.07 | 49.09 | 31.88 | 32.85 | 18.81 | 41.57 | 46.03 |