config.json

{
  "vision_config": {
    "patch_size": 16,
    "width": 768,
    "layers": 12,
    "heads": 12,
    "mlp_ratio": 4.0,
    "output_dim": 1024,
    "ls_init_value": null,
    "drop_path": 0.0,
    "image_size": 224,
    "use_abs_posemb": true,
    "use_cls_token": true,
    "use_rope2d": true,
    "pool_type": "attn",
    "attn_pooler_heads": 8,
    "use_ln_pre": true,
    "use_ln_post": true
  },
  "text_config": {
    "context_length": 32,
    "width": 1024,
    "heads": 16,
    "layers": 24,
    "output_dim": 1024,
    "mlp_ratio": 4.0,
    "vocab_size": 49408
  }
}