Mathematical Reasoning on GSM8K (exact_match)
46Exact Match AccuracyBaseline
Evaluation Results
| Method | Links | |
|---|---|---|
| BaselineBase Model=Llama 3.2 1B Instruct2026.03 | 46 | |
| Vocab. TrimmingBase Model=Llama 3.2 1B Instruct2026.03 | 46 | |
| FlashHeadBase Model=Llama 3.2 1B Instruct2026.03 | 46 | |
| SVDSoftmaxBase Model=Llama 3.2 1B Instruct2026.03 | 26 |