train/models/clip/base-16_config.json

{
  "stream_0": {
    "emb_size": 512,
    "feedforward_size": 2048,
    "hidden_size": 512,
    "hidden_act": "gelu",
    "heads_num": 8,
    "layers_num": 12,
    "max_seq_length": 512,
    "embedding": ["word", "pos"],
    "encoder": "transformer",
    "mask": "fully_visible",
    "remove_embedding_layernorm": false,
    "layernorm_positioning": "post",
    "pooling": "first"
  },

  "stream_1": {
    "emb_size": 768,
    "feedforward_size": 3072,
    "hidden_size": 768,
    "hidden_act": "gelu_fast",
    "heads_num": 12,
    "layers_num": 12,
    "max_seq_length": 197,
    "embedding": ["patch", "pos"],
    "encoder": "transformer",
    "mask": "fully_visible",
    "remove_embedding_layernorm": true,
    "layernorm_positioning": "pre",
    "pooling": "first"
  },

  "data_processor": "clip",
  "embedding": ["dual"],
  "encoder": "dual",
  "target": ["clr"],
  "image_height": 224,
  "image_width": 224,
  "patch_size": 16,
  "feature_size": 512,
  "projection": true,
  "tie_weights": false,
  "dropout": 0.0
}