torch.optim.adadelta 的源代码

```html

import torch
from torch import Tensor

from .optimizer import (Optimizer, _use_grad_for_differentiable, _default_to_fused_or_foreach,
                        _differentiable_doc, _foreach_doc, _maximize_doc, _view_as_real)
from typing import List, Optional

__all__ = ["Adadelta", "adadelta"]


class Adadelta(Optimizer):
    def __init__(
        self,
        params,
        lr=1.0,
        rho=0.9,
        eps=1e-6,
        weight_decay=0,
        foreach: Optional[bool] = None,
        *,
        maximize: bool = False,
        differentiable: bool = False,
    ):
        if not 0.0 <= lr:
            raise ValueError(f"Invalid learning rate: {lr}")
        if not 0.0 <= rho <= 1.0:
            raise ValueError(f"Invalid rho value: {rho}")
        if not 0.0 <= eps:
            raise ValueError(f"Invalid epsilon value: {eps}")
        if not 0.0 <= weight_decay:
            raise ValueError(f"Invalid weight_decay value: {weight_decay}")

        defaults = dict(
            lr=lr,
            rho=rho,
            eps=eps,
            weight_decay=weight_decay,
            maximize=maximize,
            foreach=foreach,
            differentiable=differentiable,
        )
        super().__init__(params, defaults)

    def __setstate__(self, state):
        super().__setstate__(state)
        for group in self.param_groups:
            group.setdefault("foreach", None)
            group.setdefault("maximize", False)
            group.setdefault("differentiable", False)

    def _init_group(self, group, params_with_grad, grads, square_avgs, acc_deltas):
        has_complex = False
        for p in group["params"]:
            if p.grad is None:
                continue
            has_complex |= torch.is_complex(p)
            params_with_grad.append(p)
            if p.grad.is_sparse:
                raise RuntimeError("Adadelta does not support sparse gradients")
            grads.append(p.grad)

            state = self.state[p]

            # 延迟状态初始化
            if len(state) == 0:
                state["step"] = 0
                state["square_avg"] = torch.zeros_like(
                    p, memory_format=torch.preserve_format
                )
                state["acc_delta"] = torch.zeros_like(
                    p, memory_format=torch.preserve_format
                )

            square_avgs.append(state["square_avg"])
            acc_deltas.append(state["acc_delta"])

            state["step"] += 1
        return has_complex

    @_use_grad_for_differentiable
    def step(self, closure=None):
        """执行单个优化步骤。

        参数:
            closure (Callable, 可选): 一个重新评估模型并返回损失的闭包。
        """
        loss = None
        if closure is not None<span