config.json

{
  "model_type": "tianmu_mere",
  "model_name": "Tianmu-MERE",
  "architecture": "dual_tower_retrieval_embedding",
  "vision_backbone": "SigLIP2-SO400M-patch14-384",
  "text_backbone": "BGE-base-zh-v1.5",
  "checkpoint_file": "model.safetensors",
  "vision_config_path": "vision_encoder",
  "text_config_path": "text_encoder",
  "processor_path": "processor",
  "image_size": 384,
  "embed_dim": 512,
  "normalize_embeddings": true,
  "training_teacher": "DINOv3 dense distillation was used during training only; DINO is not required for inference.",
  "state_dict_prefixes": [
    "image_proj",
    "log_inv_t",
    "text",
    "text_proj",
    "vision"
  ],
  "processor_files": [
    "preprocessor_config.json",
    "special_tokens_map.json",
    "tokenizer.json",
    "tokenizer_config.json",
    "vocab.txt"
  ]
}