Question Answering on MMLU-Sci
39.6AccuracyTaDA
Evaluation Results
| Method | Links | |
|---|---|---|
| TaDABackbone=Llama-2-7b, Number of samples=5002026.06 | 39.6 | |
| Domain onlyBackbone=Llama-2-7b, Number of samples=5002026.06 | 36.8 | |
| DARE-TIESBackbone=Llama-2-7b, Number of samples=500, Evaluation protocol=mean over 3 random seeds2026.06 | 33.3 | |
| Task onlyBackbone=Llama-2-7b, Number of samples=5002026.06 | 32.6 | |
| TIESBackbone=Llama-2-7b, Number of samples=5002026.06 | 32.6 | |
| Task Arith.Backbone=Llama-2-7b, Number of samples=5002026.06 | 31.9 | |
| Mag. PruningBackbone=Llama-2-7b, Number of samples=5002026.06 | 31.3 | |
| LinearBackbone=Llama-2-7b, Number of samples=5002026.06 | 30.6 | |
| DARE-LinBackbone=Llama-2-7b, Number of samples=500, Evaluation protocol=mean over 3 random seeds2026.06 | 30.6 | |
| SVD MergeBackbone=Llama-2-7b, Number of samples=5002026.06 | 29.9 | |
| Base modelBackbone=Llama-2-7b, Number of samples=5002026.06 | 27.8 |