Open-form mathematical reasoning on DeepMind MATHEMATICS (Exact-match accuracy)
42.13Exact-match AccuracyTRIM
Evaluation results
| Method | Links | |
|---|---|---|
| TRIMBackbone=LLAMA-2-7B, Coreset Method=TRIM, Coreset Fraction=50%2025.10 | 42.13 | |
| Full-data Fine-tuningBackbone=LLAMA-2-7B, Coreset Method=Full-data, Coreset Fraction=100%2025.10 | 42.1 | |
| S2LBackbone=LLAMA-2-7B, Coreset Method=S2L, Coreset Fraction=50%2025.10 | 41.92 | |
| LESSBackbone=LLAMA-2-7B, Coreset Method=LESS, Coreset Fraction=50%2025.10 | 41.8 | |
| TAGCOSBackbone=LLAMA-2-7B, Coreset Method=TAGCOS, Coreset Fraction=50%2025.10 | 40.84 | |
| RandomBackbone=LLAMA-2-7B, Coreset Method=Random, Coreset Fraction=50%2025.10 | 38.75 | |
| Pretrained (no Fine-tuning)Backbone=LLAMA-2-7B, Coreset Method=Pretrained2025.10 | 8.3 |