reference/code/Fast-dLLM/third_party/lmflow/optim/yogi.py

#!/usr/bin/env python
# -*- coding: utf-8 -*-

import math
import torch
import torch.nn as nn
from torch.optim.optimizer import Optimizer

class Yogi(Optimizer):
    r"""Implements Yogi Optimizer Algorithm.
    It has been proposed in `Adaptive methods for Nonconvex Optimization`.

    https://papers.nips.cc/paper/8186-adaptive-methods-for-nonconvex-optimization  # noqa

    Note:
        Reference code: https://github.com/4rtemi5/Yogi-Optimizer_Keras
    """

    def __init__(
        self,
        params,
        lr: float = 1e-2,
        betas = (0.9, 0.999),
        eps: float = 1e-3,
        initial_accumulator: float = 1e-6,
        weight_decay: float = 0,
    ) -> None:
        if lr <= 0.0:
            raise ValueError("Invalid learning rate: {}".format(lr))
        if eps < 0.0:
            raise ValueError("Invalid epsilon value: {}".format(eps))
        if not 0.0 <= betas[0] < 1.0:
            raise ValueError(
                "Invalid beta parameter at index 0: {}".format(betas[0])
            )
        if not 0.0 <= betas[1] < 1.0:
            raise ValueError(
                "Invalid beta parameter at index 1: {}".format(betas[1])
            )
        if weight_decay < 0:
            raise ValueError(
                "Invalid weight_decay value: {}".format(weight_decay)
            )

        defaults = dict(
            lr=lr,
            betas=betas,
            eps=eps,
            initial_accumulator=initial_accumulator,
            weight_decay=weight_decay,
        )
        super(Yogi, self).__init__(params, defaults)

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

        Arguments:
            closure: 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(
                        "Yogi does not support sparse gradients, "
                        "please consider SparseAdam instead"
                    )

                state = self.state[p]

                # State initialization
                # Followed from official implementation in tensorflow addons:
                # https://github.com/tensorflow/addons/blob/master/tensorflow_addons/optimizers/yogi.py#L118 # noqa
                # For more details refer to the discussion:
                # https://github.com/jettify/pytorch-optimizer/issues/77
                if len(state) == 0:
                    state["step"] = 0
                    # Exponential moving average of gradient values
                    state["exp_avg"] = nn.init.constant_(
                        torch.empty_like(
                            p.data, memory_format=torch.preserve_format
                        ),
                        group["initial_accumulator"],
                    )
                    # Exponential moving average of squared gradient values
                    state["exp_avg_sq"] = nn.init.constant_(
                        torch.empty_like(
                            p.data, memory_format=torch.preserve_format
                        ),
                        group["initial_accumulator"],
                    )

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

                state["step"] += 1
                bias_correction1 = 1 - beta1 ** state["step"]
                bias_correction2 = 1 - beta2 ** state["step"]

                if group["weight_decay"] != 0:
                    grad = grad.add(p.data, alpha=group["weight_decay"])

                # Decay the first and second moment running average coefficient
                exp_avg.mul_(beta1).add_(grad, alpha=1 - beta1)

                grad_squared = grad.mul(grad)

                exp_avg_sq.addcmul_(
                    torch.sign(exp_avg_sq - grad_squared),
                    grad_squared,
                    value=-(1 - beta2),
                )

                denom = (exp_avg_sq.sqrt() / math.sqrt(bias_correction2)).add_(
                    group["eps"]
                )
                step_size = group["lr"] / bias_correction1
                p.data.addcdiv_(exp_avg, denom, value=-step_size)

        return loss