config.json

{
  "architectures": [
    "RaonVEModel"
  ],
  "model_type": "raon_ve",
  "auto_map": {
    "AutoConfig": "configuration_raonve.RaonVEConfig",
    "AutoModel": "modeling_raonve.RaonVEModel"
  },
  "embed_dim": 1152,
  "init_logit_bias": -10,
  "vision_config": {
    "image_size": 256,
    "timm_model_name": "vit_so400m_patch16_siglip_256",
    "timm_model_pretrained": false,
    "timm_pool": "map",
    "timm_proj": "none"
  },
  "text_config": {
    "context_length": 64,
    "vocab_size": 256000,
    "hf_tokenizer_name": "timm/ViT-SO400M-16-SigLIP2-256",
    "tokenizer_kwargs": {
      "clean": "canonicalize"
    },
    "width": 1152,
    "heads": 16,
    "layers": 27,
    "mlp_ratio": 3.7362,
    "no_causal_mask": true,
    "proj_bias": true,
    "pool_type": "last",
    "norm_kwargs": {
      "eps": 1e-06
    },
    "act_kwargs": {
      "approximate": "tanh"
    }
  }
}