code/adamw.py

import math

import torch
from torch.optim import Optimizer

# From: https://github.com/uds-lsv/bert-stable-fine-tuning
# Paper: On the Stability of Fine-tuning BERT: Misconceptions, Explanations, and Strong Baselines
# Marius Mosbach, Maksym Andriushchenko, Dietrich Klakow
# https://arxiv.org/abs/2006.04884

class mAdamW(Optimizer):
    """ Implements Adam algorithm with weight decay fix.
    Parameters:
        lr (float): learning rate. Default 1e-3.
        betas (tuple of 2 floats): Adams beta parameters (b1, b2). Default: (0.9, 0.999)
        eps (float): Adams epsilon. Default: 1e-6
        weight_decay (float): Weight decay. Default: 0.0
        correct_bias (bool): can be set to False to avoid correcting bias in Adam (e.g. like in Bert TF repository). Default True.
        local_normalization (bool):
        max_grad_norm (bool):
    """

    def __init__(self, params, lr=1e-3, betas=(0.9, 0.999), eps=1e-6, weight_decay=0.0, correct_bias=True,
                 local_normalization=False, max_grad_norm=-1):
        if lr < 0.0:
            raise ValueError("Invalid learning rate: {} - should be >= 0.0".format(lr))
        if not 0.0 <= betas[0] < 1.0:
            raise ValueError("Invalid beta parameter: {} - should be in [0.0, 1.0[".format(betas[0]))
        if not 0.0 <= betas[1] < 1.0:
            raise ValueError("Invalid beta parameter: {} - should be in [0.0, 1.0[".format(betas[1]))
        if not 0.0 <= eps:
            raise ValueError("Invalid epsilon value: {} - should be >= 0.0".format(eps))
        defaults = dict(lr=lr, betas=betas, eps=eps, weight_decay=weight_decay, correct_bias=correct_bias,
                        local_normalization=local_normalization, max_grad_norm=max_grad_norm)
        super().__init__(params, defaults)

    def step(self, closure=None):
        """Performs a single optimization step.
        Arguments:
            closure (callable, optional): A closure that reevaluates the model
                and returns the loss.
        """
        loss = None
        if closure is not None:
            loss = closure()

        updates = []

        for group in self.param_groups:

            group_updates = []

            for p in group["params"]:
                if p.grad is None:
                    continue

                # Normalize gradients locally (layer-wise)
                if group["local_normalization"]:
                    torch.nn.utils.clip_grad_norm_(p, group["max_grad_norm"])

                grad = p.grad.data
                if grad.is_sparse:
                    raise RuntimeError("Adam does not support sparse gradients, please consider SparseAdam instead")

                state = self.state[p]

                # State initialization
                if len(state) == 0:
                    state["step"] = 0
                    # Exponential moving average of gradient values
                    state["exp_avg"] = torch.zeros_like(p.data)
                    # Exponential moving average of squared gradient values
                    state["exp_avg_sq"] = torch.zeros_like(p.data)

                exp_avg, exp_avg_sq = state["exp_avg"], state["exp_avg_sq"]
                beta1, beta2 = group["betas"]

                state["step"] += 1

                # Decay the first and second moment running average coefficient
                # In-place operations to update the averages at the same time
                exp_avg.mul_(beta1).add_(grad, alpha=1.0 - beta1)
                exp_avg_sq.mul_(beta2).addcmul_(grad, grad, value=1.0 - beta2)
                denom = exp_avg_sq.sqrt().add_(group["eps"])

                step_size = group["lr"]
                if group["correct_bias"]:  # No bias correction for Bert
                    bias_correction1 = 1.0 - beta1 ** state["step"]
                    bias_correction2 = 1.0 - beta2 ** state["step"]
                    step_size = step_size * math.sqrt(bias_correction2) / bias_correction1

                p.data.addcdiv_(-step_size, exp_avg, denom)

                # Just adding the square of the weights to the loss function is *not*
                # the correct way of using L2 regularization/weight decay with Adam,
                # since that will interact with the m and v parameters in strange ways.
                #
                # Instead we want to decay the weights in a manner that doesn't interact
                # with the m/v parameters. This is equivalent to adding the square
                # of the weights to the loss with plain (non-momentum) SGD.
                # Add weight decay at the end (fixed version)
                if group["weight_decay"] > 0.0:
                    p.data.add_(p.data, alpha=-group["lr"] * group["weight_decay"])

                group_updates.append((exp_avg, denom))

            updates.append(group_updates)

        return loss, updates