Aluode/PerceptionLabPortable

1# coding=utf-82# Copyright 2022 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 16 17from ...configuration_utils import PretrainedConfig18from ...utils import logging19 20 21logger = logging.get_logger(__name__)22 23 24class GitVisionConfig(PretrainedConfig):25    r"""26    This is the configuration class to store the configuration of a [`GitVisionModel`]. It is used to instantiate a GIT27    vision encoder according to the specified arguments, defining the model architecture. Instantiating a configuration28    with the defaults will yield a similar configuration to that of the vision encoder of the GIT29    [microsoft/git-base](https://huggingface.co/microsoft/git-base) 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        hidden_size (`int`, *optional*, defaults to 768):36            Dimensionality of the encoder layers and the pooler layer.37        intermediate_size (`int`, *optional*, defaults to 3072):38            Dimensionality of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.39        num_hidden_layers (`int`, *optional*, defaults to 12):40            Number of hidden layers in the Transformer encoder.41        num_attention_heads (`int`, *optional*, defaults to 12):42            Number of attention heads for each attention layer in the Transformer encoder.43        image_size (`int`, *optional*, defaults to 224):44            The size (resolution) of each image.45        patch_size (`int`, *optional*, defaults to 16):46            The size (resolution) of each patch.47        hidden_act (`str` or `function`, *optional*, defaults to `"quick_gelu"`):48            The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`,49            `"relu"`, `"selu"` and `"gelu_new"` `"quick_gelu"` are supported.50        layer_norm_eps (`float`, *optional*, defaults to 1e-5):51            The epsilon used by the layer normalization layers.52        attention_dropout (`float`, *optional*, defaults to 0.0):53            The dropout ratio for the attention probabilities.54        initializer_range (`float`, *optional*, defaults to 0.02):55            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.56 57    Example:58 59    ```python60    >>> from transformers import GitVisionConfig, GitVisionModel61 62    >>> # Initializing a GitVisionConfig with microsoft/git-base style configuration63    >>> configuration = GitVisionConfig()64 65    >>> # Initializing a GitVisionModel (with random weights) from the microsoft/git-base style configuration66    >>> model = GitVisionModel(configuration)67 68    >>> # Accessing the model configuration69    >>> configuration = model.config70    ```"""71 72    model_type = "git_vision_model"73    base_config_key = "vision_config"74 75    def __init__(76        self,77        hidden_size=768,78        intermediate_size=3072,79        num_hidden_layers=12,80        num_attention_heads=12,81        num_channels=3,82        image_size=224,83        patch_size=16,84        hidden_act="quick_gelu",85        layer_norm_eps=1e-5,86        attention_dropout=0.0,87        initializer_range=0.02,88        **kwargs,89    ):90        super().__init__(**kwargs)91 92        self.hidden_size = hidden_size93        self.intermediate_size = intermediate_size94        self.num_hidden_layers = num_hidden_layers95        self.num_attention_heads = num_attention_heads96        self.num_channels = num_channels97        self.patch_size = patch_size98        self.image_size = image_size99        self.initializer_range = initializer_range100        self.attention_dropout = attention_dropout101        self.layer_norm_eps = layer_norm_eps102        self.hidden_act = hidden_act103 104 105class GitConfig(PretrainedConfig):106    r"""107    This is the configuration class to store the configuration of a [`GitModel`]. It is used to instantiate a GIT model108    according to the specified arguments, defining the model architecture. Instantiating a configuration with the109    defaults will yield a similar configuration to that of the GIT110    [microsoft/git-base](https://huggingface.co/microsoft/git-base) architecture.111 112    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the113    documentation from [`PretrainedConfig`] for more information.114 115    Args:116        vision_config (`dict`, *optional*):117            Dictionary of configuration options used to initialize [`GitVisionConfig`].118        vocab_size (`int`, *optional*, defaults to 30522):119            Vocabulary size of the GIT model. Defines the number of different tokens that can be represented by the120            `inputs_ids` passed when calling [`GitModel`].121        hidden_size (`int`, *optional*, defaults to 768):122            Dimensionality of the encoder layers and the pooler layer.123        num_hidden_layers (`int`, *optional*, defaults to 6):124            Number of hidden layers in the Transformer encoder.125        num_attention_heads (`int`, *optional*, defaults to 12):126            Number of attention heads for each attention layer in the Transformer encoder.127        intermediate_size (`int`, *optional*, defaults to 3072):128            Dimensionality of the "intermediate" (often named feed-forward) layer in the Transformer encoder.129        hidden_act (`str` or `Callable`, *optional*, defaults to `"gelu"`):130            The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`,131            `"relu"`, `"silu"` and `"gelu_new"` are supported.132        hidden_dropout_prob (`float`, *optional*, defaults to 0.1):133            The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.134        attention_probs_dropout_prob (`float`, *optional*, defaults to 0.1):135            The dropout ratio for the attention probabilities.136        max_position_embeddings (`int`, *optional*, defaults to 1024):137            The maximum sequence length that this model might ever be used with. Typically set this to something large138            just in case (e.g., 512 or 1024 or 2048).139        initializer_range (`float`, *optional*, defaults to 0.02):140            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.141        layer_norm_eps (`float`, *optional*, defaults to 1e-12):142            The epsilon used by the layer normalization layers.143        position_embedding_type (`str`, *optional*, defaults to `"absolute"`):144            Type of position embedding. Choose one of `"absolute"`, `"relative_key"`, `"relative_key_query"`. For145            positional embeddings use `"absolute"`. For more information on `"relative_key"`, please refer to146            [Self-Attention with Relative Position Representations (Shaw et al.)](https://huggingface.co/papers/1803.02155).147            For more information on `"relative_key_query"`, please refer to *Method 4* in [Improve Transformer Models148            with Better Relative Position Embeddings (Huang et al.)](https://huggingface.co/papers/2009.13658).149        use_cache (`bool`, *optional*, defaults to `True`):150            Whether or not the model should return the last key/values attentions (not used by all models).151        num_image_with_embedding (`int`, *optional*):152            The number of temporal embeddings to add, in case the model is used for video captioning/VQA.153 154    Examples:155 156    ```python157    >>> from transformers import GitConfig, GitModel158 159    >>> # Initializing a GIT microsoft/git-base style configuration160    >>> configuration = GitConfig()161 162    >>> # Initializing a model (with random weights) from the microsoft/git-base style configuration163    >>> model = GitModel(configuration)164 165    >>> # Accessing the model configuration166    >>> configuration = model.config167    ```"""168 169    model_type = "git"170    sub_configs = {"vision_config": GitVisionConfig}171 172    def __init__(173        self,174        vision_config=None,175        vocab_size=30522,176        hidden_size=768,177        num_hidden_layers=6,178        num_attention_heads=12,179        intermediate_size=3072,180        hidden_act="gelu",181        hidden_dropout_prob=0.1,182        attention_probs_dropout_prob=0.1,183        max_position_embeddings=1024,184        initializer_range=0.02,185        layer_norm_eps=1e-12,186        pad_token_id=0,187        position_embedding_type="absolute",188        use_cache=True,189        tie_word_embeddings=False,190        bos_token_id=101,191        eos_token_id=102,192        num_image_with_embedding=None,193        **kwargs,194    ):195        super().__init__(bos_token_id=bos_token_id, eos_token_id=eos_token_id, pad_token_id=pad_token_id, **kwargs)196 197        if vision_config is None:198            vision_config = {}199            logger.info("vision_config is None. initializing the GitVisionConfig with default values.")200 201        self.vision_config = GitVisionConfig(**vision_config)202        self.vocab_size = vocab_size203        self.hidden_size = hidden_size204        self.num_hidden_layers = num_hidden_layers205        self.num_attention_heads = num_attention_heads206        self.hidden_act = hidden_act207        self.intermediate_size = intermediate_size208        self.hidden_dropout_prob = hidden_dropout_prob209        self.attention_probs_dropout_prob = attention_probs_dropout_prob210        self.max_position_embeddings = max_position_embeddings211        self.initializer_range = initializer_range212        self.layer_norm_eps = layer_norm_eps213        self.position_embedding_type = position_embedding_type214        self.use_cache = use_cache215        self.tie_word_embeddings = tie_word_embeddings216        self.num_image_with_embedding = num_image_with_embedding217 218        self.bos_token_id = bos_token_id219        self.eos_token_id = eos_token_id220 221 222__all__ = ["GitConfig", "GitVisionConfig"]223