Open-form mathematical reasoning on GSM8K
52.45AccuracyS2L
Evaluation Results
| Method | Links | |
|---|---|---|
| S2LBackbone=LLAMA-2-7B, Coreset Method=S2L, Coreset Fraction=50%2025.10 | 52.45 | |
| TRIMBackbone=LLAMA-2-7B, Coreset Method=TRIM, Coreset Fraction=50%2025.10 | 52.23 | |
| LESSBackbone=LLAMA-2-7B, Coreset Method=LESS, Coreset Fraction=50%2025.10 | 52.1 | |
| Full-data Fine-tuningBackbone=LLAMA-2-7B, Coreset Method=Full-data, Coreset Fraction=100%2025.10 | 51.15 | |
| TAGCOSBackbone=LLAMA-2-7B, Coreset Method=TAGCOS, Coreset Fraction=50%2025.10 | 50.2 | |
| RandomBackbone=LLAMA-2-7B, Coreset Method=Random, Coreset Fraction=50%2025.10 | 48.16 | |
| Pretrained (no Fine-tuning)Backbone=LLAMA-2-7B, Coreset Method=Pretrained2025.10 | 3.1 |