metadata
library_name: keras-hub
This is a PaliGemma model uploaded using the KerasHub library and can be used with JAX, TensorFlow, and PyTorch backends.
Model config:
- name: pali_gemma_backbone
- trainable: True
- vocabulary_size: 257152
- image_size: 896
- num_layers: 26
- num_query_heads: 8
- num_key_value_heads: 4
- hidden_dim: 2304
- intermediate_dim: 18432
- head_dim: 256
- vit_patch_size: 14
- vit_num_heads: 16
- vit_hidden_dim: 1152
- vit_num_layers: 27
- vit_intermediate_dim: 4304
- vit_pooling: None
- vit_classifier_activation: None
- vit_name: None
- query_head_dim_normalize: True
- use_post_ffw_norm: True
- use_post_attention_norm: True
- final_logit_soft_cap: 30
- attention_logit_soft_cap: 50
- sliding_window_size: 4096
- use_sliding_window_attention: True
- layer_norm_epsilon: 1e-06
- dropout: 0
This model card has been generated automatically and should be completed by the model author. See Model Cards documentation for more information.