def get_bert_optim()

in models.py [0:0]

• 24 lines of code
• 4 McCabe index (conditional complexity)

def get_bert_optim(network, lr, weight_decay):
    no_decay = ["bias", "LayerNorm.weight"]
    decay_params = []
    nodecay_params = []
    for n, p in network.named_parameters():
        if any(nd in n for nd in no_decay):
            decay_params.append(p)
        else:
            nodecay_params.append(p)

    optimizer_grouped_parameters = [
        {
            "params": decay_params,
            "weight_decay": weight_decay,
        },
        {
            "params": nodecay_params,
            "weight_decay": 0.0,
        },
    ]
    optimizer = AdamW(
        optimizer_grouped_parameters,
        lr=lr,
        eps=1e-8)
    return optimizer