config.json

{
  "vocab_size": 50258,
  "embed_dim": 768,
  "num_layers": 12,
  "num_heads": 12,
  "hidden_dim": 3072,
  "context_length": 128,
  "dropout": 0.1,
  "qkv_bias": true,
  "vision_dim": 768,
  "cross_attention_pos": [
    3,
    6,
    9
  ],
  "num_visual_tokens": 64,
  "perceiver_depth": 2,
  "perceiver_heads": 8
}