Aluode/PerceptionLabPortable

1# coding=utf-82# Copyright 2023 Meta Platforms, Inc. and The HuggingFace Inc. team. All rights reserved.3#4# Licensed under the Apache License, Version 2.0 (the "License");5# you may not use this file except in compliance with the License.6# You may obtain a copy of the License at7#8#     http://www.apache.org/licenses/LICENSE-2.09#10# Unless required by applicable law or agreed to in writing, software11# distributed under the License is distributed on an "AS IS" BASIS,12# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.13# See the License for the specific language governing permissions and14# limitations under the License.15"""ConvNeXTV2 model configuration"""16 17from ...configuration_utils import PretrainedConfig18from ...utils import logging19from ...utils.backbone_utils import BackboneConfigMixin, get_aligned_output_features_output_indices20 21 22logger = logging.get_logger(__name__)23 24 25class ConvNextV2Config(BackboneConfigMixin, PretrainedConfig):26    r"""27    This is the configuration class to store the configuration of a [`ConvNextV2Model`]. It is used to instantiate an28    ConvNeXTV2 model according to the specified arguments, defining the model architecture. Instantiating a29    configuration with the defaults will yield a similar configuration to that of the ConvNeXTV230    [facebook/convnextv2-tiny-1k-224](https://huggingface.co/facebook/convnextv2-tiny-1k-224) architecture.31 32    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the33    documentation from [`PretrainedConfig`] for more information.34 35    Args:36        num_channels (`int`, *optional*, defaults to 3):37            The number of input channels.38        patch_size (`int`, *optional*, defaults to 4):39            Patch size to use in the patch embedding layer.40        num_stages (`int`, *optional*, defaults to 4):41            The number of stages in the model.42        hidden_sizes (`list[int]`, *optional*, defaults to `[96, 192, 384, 768]`):43            Dimensionality (hidden size) at each stage.44        depths (`list[int]`, *optional*, defaults to `[3, 3, 9, 3]`):45            Depth (number of blocks) for each stage.46        hidden_act (`str` or `function`, *optional*, defaults to `"gelu"`):47            The non-linear activation function (function or string) in each block. If string, `"gelu"`, `"relu"`,48            `"selu"` and `"gelu_new"` are supported.49        initializer_range (`float`, *optional*, defaults to 0.02):50            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.51        layer_norm_eps (`float`, *optional*, defaults to 1e-12):52            The epsilon used by the layer normalization layers.53        drop_path_rate (`float`, *optional*, defaults to 0.0):54            The drop rate for stochastic depth.55        image_size (`int`, *optional*, defaults to 224):56            The size (resolution) of each image.57        out_features (`list[str]`, *optional*):58            If used as backbone, list of features to output. Can be any of `"stem"`, `"stage1"`, `"stage2"`, etc.59            (depending on how many stages the model has). If unset and `out_indices` is set, will default to the60            corresponding stages. If unset and `out_indices` is unset, will default to the last stage. Must be in the61            same order as defined in the `stage_names` attribute.62        out_indices (`list[int]`, *optional*):63            If used as backbone, list of indices of features to output. Can be any of 0, 1, 2, etc. (depending on how64            many stages the model has). If unset and `out_features` is set, will default to the corresponding stages.65            If unset and `out_features` is unset, will default to the last stage. Must be in the66            same order as defined in the `stage_names` attribute.67 68    Example:69    ```python70    >>> from transformers import ConvNeXTV2Config, ConvNextV2Model71 72    >>> # Initializing a ConvNeXTV2 convnextv2-tiny-1k-224 style configuration73    >>> configuration = ConvNeXTV2Config()74 75    >>> # Initializing a model (with random weights) from the convnextv2-tiny-1k-224 style configuration76    >>> model = ConvNextV2Model(configuration)77 78    >>> # Accessing the model configuration79    >>> configuration = model.config80    ```"""81 82    model_type = "convnextv2"83 84    def __init__(85        self,86        num_channels=3,87        patch_size=4,88        num_stages=4,89        hidden_sizes=None,90        depths=None,91        hidden_act="gelu",92        initializer_range=0.02,93        layer_norm_eps=1e-12,94        drop_path_rate=0.0,95        image_size=224,96        out_features=None,97        out_indices=None,98        **kwargs,99    ):100        super().__init__(**kwargs)101 102        self.num_channels = num_channels103        self.patch_size = patch_size104        self.num_stages = num_stages105        self.hidden_sizes = [96, 192, 384, 768] if hidden_sizes is None else hidden_sizes106        self.depths = [3, 3, 9, 3] if depths is None else depths107        self.hidden_act = hidden_act108        self.initializer_range = initializer_range109        self.layer_norm_eps = layer_norm_eps110        self.drop_path_rate = drop_path_rate111        self.image_size = image_size112        self.stage_names = ["stem"] + [f"stage{idx}" for idx in range(1, len(self.depths) + 1)]113        self._out_features, self._out_indices = get_aligned_output_features_output_indices(114            out_features=out_features, out_indices=out_indices, stage_names=self.stage_names115        )116 117 118__all__ = ["ConvNextV2Config"]119