Aluode/PerceptionLabPortable

1# coding=utf-82# Copyright 2022 KAIST 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"""GLPN model configuration"""16 17from ...configuration_utils import PretrainedConfig18from ...utils import logging19 20 21logger = logging.get_logger(__name__)22 23 24class GLPNConfig(PretrainedConfig):25    r"""26    This is the configuration class to store the configuration of a [`GLPNModel`]. It is used to instantiate an GLPN27    model according to the specified arguments, defining the model architecture. Instantiating a configuration with the28    defaults will yield a similar configuration to that of the GLPN29    [vinvino02/glpn-kitti](https://huggingface.co/vinvino02/glpn-kitti) architecture.30 31    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the32    documentation from [`PretrainedConfig`] for more information.33 34    Args:35        num_channels (`int`, *optional*, defaults to 3):36            The number of input channels.37        num_encoder_blocks (`int`, *optional*, defaults to 4):38            The number of encoder blocks (i.e. stages in the Mix Transformer encoder).39        depths (`list[int]`, *optional*, defaults to `[2, 2, 2, 2]`):40            The number of layers in each encoder block.41        sr_ratios (`list[int]`, *optional*, defaults to `[8, 4, 2, 1]`):42            Sequence reduction ratios in each encoder block.43        hidden_sizes (`list[int]`, *optional*, defaults to `[32, 64, 160, 256]`):44            Dimension of each of the encoder blocks.45        patch_sizes (`list[int]`, *optional*, defaults to `[7, 3, 3, 3]`):46            Patch size before each encoder block.47        strides (`list[int]`, *optional*, defaults to `[4, 2, 2, 2]`):48            Stride before each encoder block.49        num_attention_heads (`list[int]`, *optional*, defaults to `[1, 2, 5, 8]`):50            Number of attention heads for each attention layer in each block of the Transformer encoder.51        mlp_ratios (`list[int]`, *optional*, defaults to `[4, 4, 4, 4]`):52            Ratio of the size of the hidden layer compared to the size of the input layer of the Mix FFNs in the53            encoder blocks.54        hidden_act (`str` or `function`, *optional*, defaults to `"gelu"`):55            The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`,56            `"relu"`, `"selu"` and `"gelu_new"` are supported.57        hidden_dropout_prob (`float`, *optional*, defaults to 0.0):58            The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.59        attention_probs_dropout_prob (`float`, *optional*, defaults to 0.0):60            The dropout ratio for the attention probabilities.61        initializer_range (`float`, *optional*, defaults to 0.02):62            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.63        drop_path_rate (`float`, *optional*, defaults to 0.1):64            The dropout probability for stochastic depth, used in the blocks of the Transformer encoder.65        layer_norm_eps (`float`, *optional*, defaults to 1e-06):66            The epsilon used by the layer normalization layers.67        decoder_hidden_size (`int`, *optional*, defaults to 64):68            The dimension of the decoder.69        max_depth (`int`, *optional*, defaults to 10):70            The maximum depth of the decoder.71        head_in_index (`int`, *optional*, defaults to -1):72            The index of the features to use in the head.73 74    Example:75 76    ```python77    >>> from transformers import GLPNModel, GLPNConfig78 79    >>> # Initializing a GLPN vinvino02/glpn-kitti style configuration80    >>> configuration = GLPNConfig()81 82    >>> # Initializing a model from the vinvino02/glpn-kitti style configuration83    >>> model = GLPNModel(configuration)84 85    >>> # Accessing the model configuration86    >>> configuration = model.config87    ```"""88 89    model_type = "glpn"90 91    def __init__(92        self,93        num_channels=3,94        num_encoder_blocks=4,95        depths=[2, 2, 2, 2],96        sr_ratios=[8, 4, 2, 1],97        hidden_sizes=[32, 64, 160, 256],98        patch_sizes=[7, 3, 3, 3],99        strides=[4, 2, 2, 2],100        num_attention_heads=[1, 2, 5, 8],101        mlp_ratios=[4, 4, 4, 4],102        hidden_act="gelu",103        hidden_dropout_prob=0.0,104        attention_probs_dropout_prob=0.0,105        initializer_range=0.02,106        drop_path_rate=0.1,107        layer_norm_eps=1e-6,108        decoder_hidden_size=64,109        max_depth=10,110        head_in_index=-1,111        **kwargs,112    ):113        super().__init__(**kwargs)114 115        self.num_channels = num_channels116        self.num_encoder_blocks = num_encoder_blocks117        self.depths = depths118        self.sr_ratios = sr_ratios119        self.hidden_sizes = hidden_sizes120        self.patch_sizes = patch_sizes121        self.strides = strides122        self.mlp_ratios = mlp_ratios123        self.num_attention_heads = num_attention_heads124        self.hidden_act = hidden_act125        self.hidden_dropout_prob = hidden_dropout_prob126        self.attention_probs_dropout_prob = attention_probs_dropout_prob127        self.initializer_range = initializer_range128        self.drop_path_rate = drop_path_rate129        self.layer_norm_eps = layer_norm_eps130        self.decoder_hidden_size = decoder_hidden_size131        self.max_depth = max_depth132        self.head_in_index = head_in_index133 134 135__all__ = ["GLPNConfig"]136