utils/nn/model_utils.py

import numpy as np


def print_arch(model, model_name='model'):
    print(f"| {model_name} Arch: ", model)
    num_params(model, model_name=model_name)


def num_params(model, print_out=True, model_name="model"):
    parameters = filter(lambda p: p.requires_grad, model.parameters())
    parameters = sum([np.prod(p.size()) for p in parameters]) / 1_000_000
    if print_out:
        print(f'| {model_name} Trainable Parameters: %.3fM' % parameters)
    return parameters