Instruct-tuning performance on Leaderboard
0.4159BBH ScoreTaskPGM (PMI)
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| TaskPGM (PMI)Backbone=Llama2-7B, Sample Size=50K2025.07 | 0.4159 | 0.2794 | 0.3577 | 0.0189 | 0.1967 | 0.4511 | |
| LESSBackbone=Llama2-7B, Sample Size=50K2025.07 | 0.4133 | 0.2475 | 0.3407 | 0.0136 | 0.1906 | 0.4226 | |
| TaskPGM (PMI)Backbone=Llama2-7B, Sample Size=25K2025.07 | 0.4095 | 0.2718 | 0.3561 | 0.0159 | 0.1924 | 0.4298 | |
| LESSBackbone=Llama2-7B, Sample Size=25K2025.07 | 0.4059 | 0.2412 | 0.3417 | 0.0001 | 0.1816 | 0.4157 | |
| TaskPGM (JSD)Backbone=Llama2-7B, Sample Size=50K2025.07 | 0.3624 | 0.2802 | 0.3925 | 0.0098 | 0.1927 | 0.4246 | |
| EPMBackbone=Llama2-7B, Sample Size=25K2025.07 | 0.3593 | 0.2601 | 0.3405 | 0.0151 | 0.1836 | 0.4286 | |
| RandomBackbone=Llama2-7B, Sample Size=50K2025.07 | 0.3565 | 0.2668 | 0.3581 | 0.0134 | 0.1811 | 0.377 | |
| EPMBackbone=Llama2-7B, Sample Size=50K2025.07 | 0.3532 | 0.2634 | 0.3507 | 0.0128 | 0.1799 | 0.4206 | |
| UniformBackbone=Llama2-7B, Sample Size=25K2025.07 | 0.3501 | 0.2701 | 0.3501 | 0.0151 | 0.1768 | 0.4027 | |
| RandomBackbone=Llama2-7B, Sample Size=25K2025.07 | 0.3482 | 0.2626 | 0.3465 | 0.0098 | 0.1877 | 0.3677 | |
| UniformBackbone=Llama2-7B, Sample Size=50K2025.07 | 0.348 | 0.2785 | 0.3901 | 0.0161 | 0.1895 | 0.4057 | |
| TaskPGM (JSD)Backbone=Llama2-7B, Sample Size=25K2025.07 | 0.3454 | 0.2785 | 0.3729 | 0.0151 | 0.179 | 0.4021 |