config.json

{
  "model_type": "siglip",
  "model_size": "base",
  "vision_encoder": {
    "architecture": "vit_base_patch16_224",
    "hidden_size": 768,
    "num_layers": 12,
    "num_heads": 12
  },
  "text_encoder": {
    "architecture": "gpt2",
    "hidden_size": 768,
    "num_layers": 12,
    "num_heads": 12,
    "vocab_size": 50257
  },
  "embed_dim": 512,
  "projector_width": 2048,
  "projector_depth": 4,
  "training": {
    "dataset": "DataComp-large",
    "total_steps": 200000,
    "batch_size": 4096
  }
}