vision_config.json

{
  "_attn_implementation": "flash_attention_2",
  "activation_func": "gelu_pytorch_tanh",
  "attn_bias": false,
  "init_pos_emb_height": 64,
  "init_pos_emb_time": 4,
  "init_pos_emb_width": 64,
  "linear_bias": false,
  "merge_kernel_size": [
    2,
    2
  ],
  "merge_type": "sd2_tpool",
  "mlp_type": "mlp2",
  "mm_hidden_size": 1024,
  "mm_projector_type": "patchmergerv2",
  "norm_type": "rmsnorm",
  "patch_embed_proj_bias": false,
  "patch_size": 14,
  "pos_emb_interpolation_mode": "bilinear",
  "pos_emb_type": "divided_fixed",
  "projector_hidden_act": "gelu",
  "projector_ln_eps": 1e-05,
  "qkv_hidden_size": 1536,
  "text_hidden_size": 7168,
  "vt_hidden_size": 1024,
  "vt_intermediate_size": 4096,
  "vt_num_attention_heads": 12,
  "vt_num_hidden_layers": 27
}