Mathematical Reasoning on MATH 500 (Accuracy by Context Length)
29.2Accuracy (128 tokens)EvoToken (10k FT)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| EvoToken (10k FT)NFE / Gen Len=1/22026.01 | 29.2 | 36.4 | 37.4 | 34.33 | |
| BaselineNFE / Gen Len=12026.01 | 28.8 | 34.6 | 39.4 | 34.27 | |
| EvoToken (10k FT)NFE / Gen Len=12026.01 | 28.4 | 39.6 | 41 | 36.33 | |
| FT-Base (10k FT)NFE / Gen Len=12026.01 | 26.6 | 36.2 | 40.4 | 34.4 | |
| BaselineNFE / Gen Len=1/22026.01 | 26.6 | 32.2 | 33.2 | 30.67 | |
| EvoToken (10k FT)NFE / Gen Len=1/42026.01 | 23.6 | 31 | 31.2 | 28.6 | |
| BaselineNFE / Gen Len=1/42026.01 | 23.4 | 26.6 | 29.6 | 26.53 | |
| FT-Base (10k FT)NFE / Gen Len=1/22026.01 | 22.6 | 31.2 | 34 | 29.27 | |
| FT-Base (10k FT)NFE / Gen Len=1/42026.01 | 16.2 | 19.6 | 23.2 | 19.67 |