from transformers import PretrainedConfig


class BrainOCRVisionConfig(PretrainedConfig):
    model_type = "brain_ocr_vision"

    def __init__(self, image_size=560, patch_size=14, hidden_size=1280,
                 intermediate_size=5120, num_hidden_layers=32,
                 num_attention_heads=16, channels=3, layer_norm_eps=1e-6,
                 attention_dropout=0.0, **kwargs):
        super().__init__(**kwargs)
        self.image_size = image_size
        self.patch_size = patch_size
        self.hidden_size = hidden_size
        self.intermediate_size = intermediate_size
        self.num_hidden_layers = num_hidden_layers
        self.num_attention_heads = num_attention_heads
        self.channels = channels
        self.layer_norm_eps = layer_norm_eps
        self.attention_dropout = attention_dropout


class BrainOCRConfig(PretrainedConfig):
    model_type = "brain_ocr"
    sub_configs = {"vision_config": BrainOCRVisionConfig}

    def __init__(self, **kwargs):
        vision_config = kwargs.pop("vision_config", None)
        if isinstance(vision_config, dict):
            self.vision_config = BrainOCRVisionConfig(**vision_config)
        elif isinstance(vision_config, BrainOCRVisionConfig):
            self.vision_config = vision_config
        else:
            self.vision_config = BrainOCRVisionConfig()
        super().__init__(**kwargs)