text_decoder_config.json

{
  "architecture": "OpenVision2TextDecoder",
  "fusion_style": "concat",
  "width": 1024,
  "depth": 24,
  "num_heads": 16,
  "mlp_dim": 4096,
  "vocab_size": 32000,
  "vision_width": 1280,
  "layer_norm_eps": 1e-06,
  "pad_id": 0,
  "bos_id": 1,
  "eos_id": 2,
  "weights_file": "caption_decoder.safetensors",
  "modeling_file": "modeling_openvision2_decoder.py",
  "tokenizer": "bert wordpiece (bert_base_vocab_bos_eos.txt), lowercase, [PAD]=0 [bos]=1 [eos]=2"
}