config.json

{
  "architectures": [
    "Gemma4VisionModel"
  ],
  "attention_bias": false,
  "attention_dropout": 0.1,
  "dropout": 0.1,
  "dtype": "float32",
  "head_dim": 64,
  "hidden_activation": "gelu_pytorch_tanh",
  "hidden_size": 32,
  "image_size": 20,
  "initializer_range": 0.02,
  "intermediate_size": 37,
  "is_training": true,
  "max_position_embeddings": 131072,
  "model_type": "gemma4_vision",
  "num_attention_heads": 4,
  "num_channels": 3,
  "num_hidden_layers": 2,
  "num_key_value_heads": 1,
  "patch_size": 5,
  "pooling_kernel_size": 2,
  "position_embedding_size": 10240,
  "rms_norm_eps": 1e-06,
  "rope_parameters": {
    "rope_theta": 100.0,
    "rope_type": "default"
  },
  "standardize": false,
  "transformers_version": "5.16.0.dev0",
  "use_clipped_linears": false,
  "use_labels": true
}