Aluode/PerceptionLabPortable

1#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨2#           This file was automatically generated from src/transformers/models/got_ocr2/modular_got_ocr2.py.3#               Do NOT edit this file manually as any edits will be overwritten by the generation of4#             the file from the modular. If any change should be done, please apply the change to the5#                          modular_got_ocr2.py file directly. One of our CI enforces this.6#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨7# coding=utf-88# Copyright 2024 HuggingFace Inc. team. All rights reserved.9#10# Licensed under the Apache License, Version 2.0 (the "License");11# you may not use this file except in compliance with the License.12# You may obtain a copy of the License at13#14#     http://www.apache.org/licenses/LICENSE-2.015#16# Unless required by applicable law or agreed to in writing, software17# distributed under the License is distributed on an "AS IS" BASIS,18# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.19# See the License for the specific language governing permissions and20# limitations under the License.21 22 23from ...configuration_utils import PretrainedConfig24from ..auto import CONFIG_MAPPING, AutoConfig25 26 27class GotOcr2VisionConfig(PretrainedConfig):28    r"""29    This is the configuration class to store the configuration of a [`GotOcr2VisionModel`]. It is used to instantiate a GOT_OCR230    vision encoder according to the specified arguments, defining the model architecture. Instantiating a configuration31    defaults will yield a similar configuration to that of the SAM ViT-h32    [facebook/sam-vit-huge](https://huggingface.co/facebook/sam-vit-huge) architecture.33 34    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the35    documentation from [`PretrainedConfig`] for more information.36 37    Args:38        hidden_size (`int`, *optional*, defaults to 768):39            Dimensionality of the encoder layers and the pooler layer.40        output_channels (`int`, *optional*, defaults to 256):41            Dimensionality of the output channels in the Patch Encoder.42        num_hidden_layers (`int`, *optional*, defaults to 12):43            Number of hidden layers in the Transformer encoder.44        num_attention_heads (`int`, *optional*, defaults to 12):45            Number of attention heads for each attention layer in the Transformer encoder.46        num_channels (`int`, *optional*, defaults to 3):47            Number of channels in the input image.48        image_size (`int`, *optional*, defaults to 1024):49            Expected resolution. Target size of the resized input image.50        patch_size (`int`, *optional*, defaults to 16):51            Size of the patches to be extracted from the input image.52        hidden_act (`str`, *optional*, defaults to `"gelu"`):53            The non-linear activation function (function or string)54        layer_norm_eps (`float`, *optional*, defaults to 1e-06):55            The epsilon used by the layer normalization layers.56        attention_dropout (`float`, *optional*, defaults to 0.0):57            The dropout ratio for the attention probabilities.58        initializer_range (`float`, *optional*, defaults to 1e-10):59            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.60        qkv_bias (`bool`, *optional*, defaults to `True`):61            Whether to add a bias to query, key, value projections.62        use_abs_pos (`bool`, *optional*, defaults to `True`):63            Whether to use absolute position embedding.64        use_rel_pos (`bool`, *optional*, defaults to `True`):65            Whether to use relative position embedding.66        window_size (`int`, *optional*, defaults to 14):67            Window size for relative position.68        global_attn_indexes (`list[int]`, *optional*, defaults to `[2, 5, 8, 11]`):69            The indexes of the global attention layers.70        mlp_dim (`int`, *optional*, defaults to 3072):71            The dimensionality of the MLP layer in the Transformer encoder.72    """73 74    base_config_key = "vision_config"75 76    def __init__(77        self,78        hidden_size=768,79        output_channels=256,80        num_hidden_layers=12,81        num_attention_heads=12,82        num_channels=3,83        image_size=1024,84        patch_size=16,85        hidden_act="gelu",86        layer_norm_eps=1e-06,87        attention_dropout=0.0,88        initializer_range=1e-10,89        qkv_bias=True,90        use_abs_pos=True,91        use_rel_pos=True,92        window_size=14,93        global_attn_indexes=[2, 5, 8, 11],94        mlp_dim=3072,95        **kwargs,96    ):97        super().__init__(**kwargs)98 99        self.hidden_size = hidden_size100        self.output_channels = output_channels101        self.num_hidden_layers = num_hidden_layers102        self.num_attention_heads = num_attention_heads103        self.num_channels = num_channels104        self.image_size = image_size105        self.patch_size = patch_size106        self.hidden_act = hidden_act107        self.layer_norm_eps = layer_norm_eps108        self.attention_dropout = attention_dropout109        self.initializer_range = initializer_range110        self.qkv_bias = qkv_bias111        self.use_abs_pos = use_abs_pos112        self.use_rel_pos = use_rel_pos113        self.window_size = window_size114        self.global_attn_indexes = global_attn_indexes115        self.mlp_dim = mlp_dim116 117 118class GotOcr2Config(PretrainedConfig):119    r"""120    This is the configuration class to store the configuration of a [`GotOcr2ForConditionalGeneration`]. It is used to instantiate a121    GotOcr2 model according to the specified arguments, defining the model architecture. Instantiating a configuration122    with the defaults will yield a similar configuration to that of GOT-OCR-2.0.123 124    e.g [stepfun-ai/GOT-OCR-2.0-hf](https://huggingface.co/stepfun-ai/GOT-OCR-2.0-hf)125 126    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the127    documentation from [`PretrainedConfig`] for more information.128 129 130    Args:131        vision_config (`Union[AutoConfig, dict]`,  *optional*, defaults to `CLIPVisionConfig`):132            The config object or dictionary of the vision backbone.133        text_config (`Union[AutoConfig, dict]`, *optional*, defaults to `LlamaConfig`):134            The config object or dictionary of the text backbone.135        image_token_index (`int`, *optional*, defaults to 151859):136            The image token index to encode the image prompt.137        image_seq_length (`int`, *optional*, defaults to 576):138            Sequence length of one image embedding.139        pad_token_id (`int`, *optional*, defaults to -1):140            Padding token id.141 142    ```python143    >>> from transformers import GotOcr2ForConditionalGeneration, GotOcr2Config144 145    >>> # Initializing a GotOcr2 style configuration146    >>> configuration = GotOcr2Config()147 148    >>> # Initializing a model from the Qwen2-VL-7B style configuration149    >>> model = GotOcr2ForConditionalGeneration(configuration)150 151    >>> # Accessing the model configuration152    >>> configuration = model.config153    ```"""154 155    model_type = "got_ocr2"156    attribute_map = {157        "image_token_id": "image_token_index",158    }159    sub_configs = {"text_config": AutoConfig, "vision_config": GotOcr2VisionConfig}160 161    def __init__(162        self,163        vision_config=None,164        text_config=None,165        image_token_index=151859,166        image_seq_length=576,167        pad_token_id=-1,168        **kwargs,169    ):170        self.image_token_index = image_token_index171        self.image_seq_length = image_seq_length172        self.pad_token_id = pad_token_id173 174        if vision_config is None:175            self.vision_config = GotOcr2VisionConfig()176        elif isinstance(vision_config, dict):177            self.vision_config = GotOcr2VisionConfig(**vision_config)178        elif isinstance(vision_config, GotOcr2VisionConfig):179            self.vision_config = vision_config180 181        if isinstance(text_config, dict):182            text_config["model_type"] = text_config.get("model_type", "qwen2")183            text_config = CONFIG_MAPPING[text_config["model_type"]](**text_config)184        elif text_config is None:185            text_config = CONFIG_MAPPING["qwen2"](186                vocab_size=151860,187                hidden_size=1024,188                intermediate_size=2816,189                num_hidden_layers=24,190                num_attention_heads=16,191                num_key_value_heads=16,192                hidden_act="silu",193                max_position_embeddings=32768,194                initializer_range=0.02,195                rms_norm_eps=1e-6,196                use_cache=True,197                tie_word_embeddings=True,198                rope_theta=1000000.0,199                rope_scaling=None,200                use_sliding_window=False,201                sliding_window=4096,202                max_window_layers=21,203                attention_dropout=0.0,204            )205 206        self.text_config = text_config207 208        super().__init__(**kwargs)209 210 211__all__ = ["GotOcr2VisionConfig", "GotOcr2Config"]212