Modular Addition on Modular Addition Token Embedding
99.4Match Accuracyauxiliary modulus Kq
Evaluation Results
| Method | Links | |
|---|---|---|
| auxiliary modulus KqN=8, q=31, Training samples=1M, Embedding=Token2026.05 | 99.4 | |
| auxiliary modulus KqN=16, q=31, Training samples=1M, Embedding=Token2026.05 | 98.9 | |
| BaselineN=8, q=31, Training samples=1M, Embedding=Token2026.05 | 98.6 | |
| auxiliary modulus KqN=32, q=31, Training samples=1M, Embedding=Token2026.05 | 96.5 | |
| auxiliary modulus KqN=8, q=97, Training samples=1M, Embedding=Token2026.05 | 90.1 | |
| auxiliary modulus KqN=8, q=31, Training samples=100K, Embedding=Token2026.05 | 89.1 | |
| auxiliary modulus KqN=16, q=97, Training samples=1M, Embedding=Token2026.05 | 81.9 | |
| BaselineN=16, q=31, Training samples=1M, Embedding=Token2026.05 | 80.7 | |
| BaselineN=8, q=97, Training samples=1M, Embedding=Token2026.05 | 74.9 | |
| auxiliary modulus KqN=32, q=97, Training samples=1M, Embedding=Token2026.05 | 62.2 | |
| auxiliary modulus KqN=16, q=31, Training samples=100K, Embedding=Token2026.05 | 58.4 | |
| BaselineN=32, q=31, Training samples=1M, Embedding=Token2026.05 | 38.7 | |
| auxiliary modulus KqN=8, q=257, Training samples=1M, Embedding=Token2026.05 | 34 | |
| auxiliary modulus KqN=32, q=31, Training samples=100K, Embedding=Token2026.05 | 33.1 | |
| auxiliary modulus KqN=16, q=257, Training samples=1M, Embedding=Token2026.05 | 21.7 | |
| auxiliary modulus KqN=8, q=97, Training samples=100K, Embedding=Token2026.05 | 13.9 | |
| auxiliary modulus KqN=32, q=257, Training samples=1M, Embedding=Token2026.05 | 12.6 | |
| auxiliary modulus KqN=16, q=97, Training samples=100K, Embedding=Token2026.05 | 6.2 | |
| BaselineN=16, q=31, Training samples=100K, Embedding=Token2026.05 | 3.3 | |
| BaselineN=8, q=31, Training samples=100K, Embedding=Token2026.05 | 3.2 | |
| BaselineN=32, q=31, Training samples=100K, Embedding=Token2026.05 | 2.9 | |
| auxiliary modulus KqN=32, q=97, Training samples=100K, Embedding=Token2026.05 | 1.7 | |
| BaselineN=16, q=97, Training samples=1M, Embedding=Token2026.05 | 1.1 | |
| BaselineN=32, q=97, Training samples=100K, Embedding=Token2026.05 | 1.1 | |
| auxiliary modulus KqN=8, q=257, Training samples=100K, Embedding=Token2026.05 | 1 | |
| BaselineN=16, q=97, Training samples=100K, Embedding=Token2026.05 | 1 | |
| BaselineN=32, q=97, Training samples=1M, Embedding=Token2026.05 | 1 | |
| BaselineN=8, q=97, Training samples=100K, Embedding=Token2026.05 | 0.9 | |
| auxiliary modulus KqN=16, q=257, Training samples=100K, Embedding=Token2026.05 | 0.6 | |
| BaselineN=8, q=257, Training samples=100K, Embedding=Token2026.05 | 0.5 | |
| BaselineN=8, q=257, Training samples=1M, Embedding=Token2026.05 | 0.5 | |
| auxiliary modulus KqN=32, q=257, Training samples=100K, Embedding=Token2026.05 | 0.5 | |
| BaselineN=16, q=257, Training samples=100K, Embedding=Token2026.05 | 0.4 | |
| BaselineN=16, q=257, Training samples=1M, Embedding=Token2026.05 | 0.4 | |
| BaselineN=32, q=257, Training samples=100K, Embedding=Token2026.05 | 0.4 | |
| BaselineN=32, q=257, Training samples=1M, Embedding=Token2026.05 | 0.3 |