DoruC/Grounded-Segment-Anything

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""" Data2VecText configuration"""16from collections import OrderedDict17from typing import Mapping18 19from ...configuration_utils import PretrainedConfig20from ...onnx import OnnxConfig21from ...utils import logging22 23 24logger = logging.get_logger(__name__)25 26DATA2VEC_TEXT_PRETRAINED_CONFIG_ARCHIVE_MAP = {27    "facebook/data2vec-text-base": "https://huggingface.co/data2vec/resolve/main/config.json",28}29 30 31class Data2VecTextConfig(PretrainedConfig):32    r"""33    This is the configuration class to store the configuration of a [`Data2VecTextModel`] and [`Data2VecTextModel`]. It34    is used to instantiate a Data2VecText model according to the specified arguments, defining the model architecture.35    Instantiating a configuration with the defaults will yield a similar configuration to that of the Data2VecText36    [facebook/data2vec-text-base](https://huggingface.co/facebook/data2vec-text-base) architecture.37 38    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the39    documentation from [`PretrainedConfig`] for more information.40 41 42    Args:43        vocab_size (`int`, *optional*, defaults to 30522):44            Vocabulary size of the DATA2VEC model. Defines the number of different tokens that can be represented by45            the `inputs_ids` passed when calling [`Data2VecModel`].46        hidden_size (`int`, *optional*, defaults to 768):47            Dimensionality of the encoder layers and the pooler layer.48        num_hidden_layers (`int`, *optional*, defaults to 12):49            Number of hidden layers in the Transformer encoder.50        num_attention_heads (`int`, *optional*, defaults to 12):51            Number of attention heads for each attention layer in the Transformer encoder.52        intermediate_size (`int`, *optional*, defaults to 3072):53            Dimensionality of the "intermediate" (often named feed-forward) layer in the Transformer encoder.54        hidden_act (`str` or `Callable`, *optional*, defaults to `"gelu"`):55            The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`,56            `"relu"`, `"silu"` and `"gelu_new"` are supported.57        hidden_dropout_prob (`float`, *optional*, defaults to 0.1):58            The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.59        attention_probs_dropout_prob (`float`, *optional*, defaults to 0.1):60            The dropout ratio for the attention probabilities.61        max_position_embeddings (`int`, *optional*, defaults to 512):62            The maximum sequence length that this model might ever be used with. Typically set this to something large63            just in case (e.g., 512 or 1024 or 2048).64        type_vocab_size (`int`, *optional*, defaults to 2):65            The vocabulary size of the `token_type_ids` passed when calling [`Data2VecModel`].66        initializer_range (`float`, *optional*, defaults to 0.02):67            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.68        layer_norm_eps (`float`, *optional*, defaults to 1e-12):69            The epsilon used by the layer normalization layers.70        position_embedding_type (`str`, *optional*, defaults to `"absolute"`):71            Type of position embedding. Choose one of `"absolute"`, `"relative_key"`, `"relative_key_query"`. For72            positional embeddings use `"absolute"`. For more information on `"relative_key"`, please refer to73            [Self-Attention with Relative Position Representations (Shaw et al.)](https://arxiv.org/abs/1803.02155).74            For more information on `"relative_key_query"`, please refer to *Method 4* in [Improve Transformer Models75            with Better Relative Position Embeddings (Huang et al.)](https://arxiv.org/abs/2009.13658).76        is_decoder (`bool`, *optional*, defaults to `False`):77            Whether the model is used as a decoder or not. If `False`, the model is used as an encoder.78        use_cache (`bool`, *optional*, defaults to `True`):79            Whether or not the model should return the last key/values attentions (not used by all models). Only80            relevant if `config.is_decoder=True`.81        classifier_dropout (`float`, *optional*):82            The dropout ratio for the classification head.83 84    Examples:85 86    ```python87    >>> from transformers import Data2VecTextConfig, Data2VecTextModel88 89    >>> # Initializing a Data2VecText facebook/data2vec-text-base style configuration90    >>> configuration = Data2VecTextConfig()91 92    >>> # Initializing a model (with random weights) from the facebook/data2vec-text-base style configuration93    >>> model = Data2VecTextModel(configuration)94 95    >>> # Accessing the model configuration96    >>> configuration = model.config97    ```"""98    model_type = "data2vec-text"99 100    def __init__(101        self,102        vocab_size=30522,103        hidden_size=768,104        num_hidden_layers=12,105        num_attention_heads=12,106        intermediate_size=3072,107        hidden_act="gelu",108        hidden_dropout_prob=0.1,109        attention_probs_dropout_prob=0.1,110        max_position_embeddings=512,111        type_vocab_size=2,112        initializer_range=0.02,113        layer_norm_eps=1e-12,114        pad_token_id=1,115        bos_token_id=0,116        eos_token_id=2,117        position_embedding_type="absolute",118        use_cache=True,119        classifier_dropout=None,120        **kwargs,121    ):122        super().__init__(pad_token_id=pad_token_id, bos_token_id=bos_token_id, eos_token_id=eos_token_id, **kwargs)123 124        self.vocab_size = vocab_size125        self.hidden_size = hidden_size126        self.num_hidden_layers = num_hidden_layers127        self.num_attention_heads = num_attention_heads128        self.hidden_act = hidden_act129        self.intermediate_size = intermediate_size130        self.hidden_dropout_prob = hidden_dropout_prob131        self.attention_probs_dropout_prob = attention_probs_dropout_prob132        self.max_position_embeddings = max_position_embeddings133        self.type_vocab_size = type_vocab_size134        self.initializer_range = initializer_range135        self.layer_norm_eps = layer_norm_eps136        self.position_embedding_type = position_embedding_type137        self.use_cache = use_cache138        self.classifier_dropout = classifier_dropout139 140 141class Data2VecTextOnnxConfig(OnnxConfig):142    @property143    def inputs(self) -> Mapping[str, Mapping[int, str]]:144        if self.task == "multiple-choice":145            dynamic_axis = {0: "batch", 1: "choice", 2: "sequence"}146        else:147            dynamic_axis = {0: "batch", 1: "sequence"}148        return OrderedDict(149            [150                ("input_ids", dynamic_axis),151                ("attention_mask", dynamic_axis),152            ]153        )154