gemma3-finetuned-hf

This is a Gemma3 model uploaded using the KerasHub library and can be used with JAX, TensorFlow, and PyTorch backends. This model is related to a CausalLM task.

Model config:

• name: gemma3_backbone
• trainable: True
• dtype: {'module': 'keras', 'class_name': 'DTypePolicy', 'config': {'name': 'float32'}, 'registered_name': None}
• vocabulary_size: 262144
• image_size: 896
• num_layers: 34
• num_query_heads: 8
• num_key_value_heads: 4
• hidden_dim: 2560
• intermediate_dim: 10240
• head_dim: 256
• query_head_dim_normalize: True
• use_query_key_norm: True
• use_post_ffw_norm: True
• use_post_attention_norm: True
• attention_logit_soft_cap: None
• final_logit_soft_cap: None
• use_sliding_window_attention: True
• sliding_window_size: 1024
• local_rope_scaling_factor: 1.0
• global_rope_scaling_factor: 8.0
• vision_encoder: {'module': 'keras_hub.src.models.gemma3.gemma3_vision_encoder', 'class_name': 'Gemma3VisionEncoder', 'config': {'name': 'gemma3_vision_encoder', 'trainable': False, 'dtype': {'module': 'keras', 'class_name': 'DTypePolicy', 'config': {'name': 'float32'}, 'registered_name': None}, 'num_heads': 16, 'hidden_dim': 1152, 'num_layers': 27, 'intermediate_dim': 4304, 'output_dim': 2560, 'pool_size': 4, 'image_size': 896, 'patch_size': 14, 'layer_norm_epsilon': 1e-06}, 'registered_name': 'keras_hub>Gemma3VisionEncoder'}
• use_bidirectional_attention: False
• layer_norm_epsilon: 1e-06
• dropout: 0
• is_embedding_model: False
• pooling_intermediate_dim: None
• embedding_dim: None

This model card has been generated automatically and should be completed by the model author. See Model Cards documentation for more information.