config.json

{
  "architectures": [
    "Kosmos2ForConditionalGeneration"
  ],
  "dtype": "float32",
  "latent_query_num": 3,
  "model_type": "kosmos-2",
  "text_config": {
    "activation_dropout": 0.0,
    "activation_function": "gelu",
    "add_cross_attention": false,
    "attention_dropout": 0.1,
    "attention_heads": 4,
    "bos_token_id": 0,
    "dropout": 0.1,
    "embed_dim": 32,
    "eos_token_id": 2,
    "ffn_dim": 37,
    "init_std": 0.02,
    "layer_norm_eps": 1e-05,
    "layerdrop": 0.0,
    "layers": 2,
    "max_position_embeddings": 512,
    "model_type": "kosmos_2_text_model",
    "pad_token_id": 1,
    "scale_embedding": true,
    "tie_word_embeddings": true,
    "use_cache": true,
    "vocab_size": 65037
  },
  "tie_word_embeddings": true,
  "transformers_version": "5.16.0.dev0",
  "vision_config": {
    "attention_dropout": 0.1,
    "dropout": 0.1,
    "hidden_act": "quick_gelu",
    "hidden_size": 32,
    "image_size": 32,
    "initializer_factor": 1.0,
    "initializer_range": 1e-10,
    "intermediate_size": 37,
    "layer_norm_eps": 1e-05,
    "model_type": "kosmos_2_vision_model",
    "num_attention_heads": 4,
    "num_channels": 3,
    "num_hidden_layers": 2,
    "patch_size": 4
  }
}