mindformers.models.clip.CLIPVisionConfig

class mindformers.models.clip.CLIPVisionConfig(hidden_size: Optional[int] = 768, intermediate_size: Optional[int] = 3072, num_hidden_layers: Optional[int] = 12, num_attention_heads: Optional[int] = 12, image_size: Optional[int] = 224, patch_size: Optional[int] = 32, hidden_act: Optional[str] = 'quick_gelu', dropout: Optional[float] = 0.0, attention_dropout: Optional[float] = 0.0, initializer_range: Optional[float] = 0.02, initializer_factor: Optional[float] = 1.0, **kwargs)

Config For CLIP Vision Module

Args:

hidden_size (Optional[int]): Dimensionality of the encoder layers and the pooler layer. intermediate_size (Optional[int]): Dimensionality of the “intermediate”

(i.e., feed-forward) layer in the Transformer encoder.

num_hidden_layers (Optional[int]): Number of hidden layers in the Transformer encoder. num_attention_heads (Optional[int]): Number of attention heads for

each attention layer in the Transformer encoder.

image_size (Optional[int]): The size (resolution) of each image. patch_size (Optional[int]): The size (resolution) of each patch. hidden_act (Optional[str]): The non-linear activation function

(function or string) in the encoder and pooler. Only “quick_gelu” supported currently.

dropout (Optional[float]): The dropout probabilitiy for all fully connected

layers in the embeddings, encoder, and pooler.

attention_dropout (Optional[float]): The dropout ratio for the attention probabilities. initializer_range (Optional[float]): The standard deviation of the

truncated_normal_initializer for initializing all weight matrices.

initializer_factor (Optional[float]): A factor for initializing all weight matrices

(should be kept to 1, used internally for initialization testing).

Examples:

>>> from mindformers import CLIPVisionConfig
>>> CLIPVisionConfig(hidden_size=512, image_size=256)
    {'hidden_size': 512, 'intermediate_size': 3072, 'num_hidden_layers': 12,
     'num_attention_heads': 16, 'image_size': 256, 'patch_size': 32,
      'hidden_act': 'quick_gelu', 'dropout': 0.0, 'attention_dropout': 0.0,
       'initializer_range': 0.02, 'initializer_factor': 1.0}