# Copyright (c) Facebook, Inc. and its affiliates. All rights reserved.

from ... import Tensor
from .module import Module
from .linear import Linear
from collections import namedtuple
from typing import List, Sequence
from .container import ModuleList

_ASMoutput = namedtuple('_ASMoutput', ['output', 'loss'])


class AdaptiveLogSoftmaxWithLoss(Module):
    in_features: int = ...
    n_classes: int = ...
    cutoffs: List[int] = ...
    div_value: float = ...
    head_bias: bool = ...
    head: Linear = ...
    tail: ModuleList = ...

    def __init__(self, in_features: int, n_classes: int, cutoffs: Sequence[int], div_value: float = ...,
                 head_bias: bool = ...) -> None: ...

    def reset_parameters(self) -> None: ...

    def forward(self, input: Tensor, target: Tensor) -> _ASMoutput: ...  # type: ignore

    def __call__(self, input: Tensor, target: Tensor) -> _ASMoutput: ...  # type: ignore

    def log_prob(self, input: Tensor) -> List[float]: ...

    def predict(self, input: Tensor) -> Tensor: ...