| { |
| "entry_class": "model.VanillaModel", |
| "output_base": 10, |
| "framework": "pytorch", |
| "model_description": "3-layer feedforward MLP (input -> 512 hidden ReLU -> output). Input: zero-padded decimal digits of a mod p, b mod p, and p concatenated into a 30-float vector. Output: 10 digit positions x 10 classes (base-10 digits of the answer, MSB first). ~16K parameters.", |
| "training_description": "Trained from random init on 200K synthetic (a, b, p) -> (a*b mod p) examples spanning tiers 1-4 (p up to 32 bits). Adam optimizer, lr=1e-3 with cosine annealing, 30 epochs, batch size 512. Best checkpoint saved by validation accuracy." |
| } |
|
|
