Modular Multiplication on Modular Multiplication p=97 (val)
910Steps to 95% ConvergenceGrokfast
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GrokfastModel=2-layer transformer2025.10 | 910 | 114 | 125.3 | 100 | |
| EGD(RSVD,r=36)Model=2-layer transformer, Implementation=Randomized SVD, rank=362025.10 | 980 | 129.3 | 131.9 | 100 | |
| EGD(SVD)Model=2-layer transformer, Implementation=Exact SVD2025.10 | 1,170 | 179.9 | 153.8 | 100 | |
| Vanilla AdamModel=2-layer transformer2025.10 | 55,780 | 5,198.7 | 93.2 | 99.1 |