config.json

{
  "architectures": [
    "Granite4VisionModel"
  ],
  "deepstack_layer_map": [
    [
      1,
      0
    ]
  ],
  "downsample_rate": "1/2",
  "dtype": "float32",
  "image_grid_pinpoints": [
    [
      8,
      8
    ]
  ],
  "image_seq_length": 576,
  "image_token_index": 3,
  "model_type": "granite4_vision",
  "pad_token_id": 32001,
  "projector_dropout": 0.0,
  "qformer_config": {
    "attention_probs_dropout_prob": 0.1,
    "cross_attention_frequency": 1,
    "encoder_hidden_size": 64,
    "hidden_act": "gelu",
    "hidden_dropout_prob": 0.1,
    "hidden_size": 64,
    "initializer_range": 0.02,
    "intermediate_size": 64,
    "layer_norm_eps": 1e-12,
    "max_position_embeddings": 2048,
    "model_type": "blip_2_qformer",
    "num_attention_heads": 1,
    "num_hidden_layers": 1,
    "pad_token_id": 0,
    "use_qformer_text_input": false,
    "vocab_size": 30522
  },
  "spatial_target_layers": [
    12,
    15,
    18,
    21
  ],
  "spatial_vision_layer": -1,
  "text_config": {
    "attention_bias": false,
    "attention_dropout": 0.0,
    "attention_multiplier": 1.0,
    "bos_token_id": 1,
    "embedding_multiplier": 1.0,
    "eos_token_id": 2,
    "hidden_act": "gelu",
    "hidden_size": 64,
    "initializer_range": 0.02,
    "intermediate_size": 64,
    "logits_scaling": 1.0,
    "max_position_embeddings": 512,
    "mlp_bias": false,
    "model_type": "granite",
    "num_attention_heads": 2,
    "num_hidden_layers": 2,
    "num_key_value_heads": 2,
    "pad_token_id": 32001,
    "residual_multiplier": 1.0,
    "rms_norm_eps": 1e-06,
    "rope_parameters": {
      "rope_theta": 10000.0,
      "rope_type": "default"
    },
    "tie_word_embeddings": false,
    "use_cache": true,
    "vocab_size": 32002
  },
  "tie_word_embeddings": false,
  "transformers_version": "5.16.0.dev0",
  "vision_config": {
    "attention_dropout": 0.0,
    "hidden_act": "gelu",
    "hidden_size": 64,
    "image_size": 8,
    "initializer_factor": 1.0,
    "initializer_range": 0.02,
    "intermediate_size": 64,
    "layer_norm_eps": 1e-05,
    "model_type": "clip_vision_model",
    "num_attention_heads": 2,
    "num_channels": 3,
    "num_hidden_layers": 2,
    "patch_size": 2,
    "projection_dim": 64
  },
  "vision_feature_layer": -1,
  "vision_feature_select_strategy": "default"
}