metadata
library_name: keras-hub
This is a SigLIP model uploaded using the KerasHub library and can be used with JAX, TensorFlow, and PyTorch backends.
Model config:
- name: sig_lip_backbone
- trainable: True
- vision_encoder: {'module': 'keras_hub.src.models.siglip.siglip_vision_encoder', 'class_name': 'SigLIPVisionEncoder', 'config': {'name': 'sig_lip_vision_encoder', 'trainable': True, 'patch_size': 16, 'hidden_dim': 1536, 'num_layers': 40, 'num_heads': 16, 'intermediate_dim': 6144, 'intermediate_activation': 'gelu_approximate', 'layer_norm_epsilon': 1e-06, 'image_shape': [256, 256, 3]}, 'registered_name': 'keras_hub>SigLIPVisionEncoder'}
- text_encoder: {'module': 'keras_hub.src.models.siglip.siglip_text_encoder', 'class_name': 'SigLIPTextEncoder', 'config': {'name': 'sig_lip_text_encoder', 'trainable': True, 'vocabulary_size': 256000, 'embedding_dim': 1152, 'hidden_dim': 1152, 'num_layers': 27, 'num_heads': 16, 'intermediate_dim': 4304, 'intermediate_activation': 'gelu_approximate', 'layer_norm_epsilon': 1e-06, 'max_sequence_length': 64}, 'registered_name': 'keras_hub>SigLIPTextEncoder'}
This model card has been generated automatically and should be completed by the model author. See Model Cards documentation for more information.