PhiNet on MNIST
This repository contains checkpoints for the MNIST dataset for the following combinations of PhiNet's hyperparameters:
| Model | Top 1 Accuracy | Top 5 Accuracy |
|---|---|---|
PhiNet(alpha=0.5, beta=1, t_zero=6, num_layers=4) |
99.12% | 100% |
PhiNet(alpha=0.75, beta=1, t_zero=6, num_layers=5) |
98.94% | 99.98% |
To download and use this repo:
from micromind import PhiNet
model = PhiNet.from_pretrained("MNIST", alpha=0.5, beta=1.0, t_zero=6, num_layers=4, num_classes=10, resolution=28)