config.json

{
  "_name_or_path": "./polyverse_v2_e1",
  "architectures": [
    "PolyverseForConditionalGeneration"
  ],
  "hidden_size": 768,
  "image_token_index": 256204,
  "is_encoder_decoder": true,
  "model_type": "Polyverse",
  "projection_dim": 1024,
  "sampler_tokens": 64,
  "text_config": {
    "_name_or_path": "facebook/nllb-200-distilled-600M",
    "architectures": [
      "M2M100ForConditionalGeneration"
    ],
    "decoder_layerdrop": 0,
    "encoder_layerdrop": 0,
    "max_length": 200,
    "model_type": "m2m_100",
    "num_image_tokens": 256,
    "tokenizer_class": "NllbTokenizer",
    "torch_dtype": "float32",
    "vocab_size": 256206
  },
  "torch_dtype": "float32",
  "transformers_version": "4.44.0",
  "vision_config": {
    "_name_or_path": "google/siglip-base-patch16-256",
    "image_size": 256,
    "model_type": "siglip_vision_model",
    "projection_dim": 1024
  }
}