ChildTuningOptimizer.py

# original code credit - https://github.com/RunxinXu/ChildTuning
# this is from the paper Child Tuning: https://arxiv.org/pdf/2109.05687.pdf


import torch
from torch.optim import Optimizer
from typing import Callable, Iterable, Tuple
from torch.distributions.bernoulli import Bernoulli
import math


class ChildTuningAdamW(Optimizer):
    def __init__(
        self,
        params: Iterable[torch.nn.parameter.Parameter],
        lr: float = 1e-3,
        betas: Tuple[float, float] = (0.9, 0.999),
        eps: float = 1e-8,
        weight_decay: float = 0.0,
        correct_bias: bool = True,
        reserve_p=0.5,
        mode=None,
    ):
        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,
        )
        super().__init__(params, defaults)

        self.gradient_mask = None
        self.reserve_p = reserve_p
        self.mode = mode
        if mode == "taskfree":
            print(f"--> optimizer running with taskfree Child Tuning")
        elif mode == "task":
            print(f"--> optimizer running with fim masking active")
        else:
            print(
                f"--> WARNING: No mode Set!  This is the same as running regular AdamW"
            )

    def set_gradient_mask(self, gradient_mask):
        self.gradient_mask = gradient_mask

    def step(self, closure: Callable = None):
        """
        Performs a single optimization step.

        Arguments:
            closure (:obj:`Callable`, `optional`): A closure that reevaluates the model and returns the loss.
        """
        loss = None
        if closure is not None:
            loss = closure()
        for group in self.param_groups:
            for p in group["params"]:
                if p.grad is None:
                    continue
                grad = p.grad.data
                if grad.is_sparse:
                    raise RuntimeError(
                        "Adam does not support sparse gradients, please consider SparseAdam instead"
                    )

                # =================== Child Tuning BEGIN =======================
                if self.mode is not None:
                    if self.mode == "task":
                        if p in self.gradient_mask:
                            grad *= self.gradient_mask[p]
                    elif self.mode == "taskfree":
                        # ChildTuning-F
                        grad_mask = Bernoulli(
                            grad.new_full(size=grad.size(), fill_value=self.reserve_p)
                        )
                        grad *= grad_mask.sample() / self.reserve_p
                    else:
                        raise ValueError(
                            "running Child Tuning optimizer but no mode set...aborting"
                        )
                # =================== Child Tuning END =======================

                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_(exp_avg, denom, value=-step_size)

                # 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)
                p.data.add_(p.data, alpha=-group["lr"] * group["weight_decay"])

        return loss