blip3o/model/multimodal_encoder/builder.py

from .ta_tok_encoder import TATokVisionTower
import torch

def build_vision_tower(vision_tower_cfg, **kwargs):
    vision_tower = getattr(vision_tower_cfg, "mm_vision_tower", getattr(vision_tower_cfg, "vision_tower", None))
    # return TATok by default, you can add more tokenizers here   
    return TATokVisionTower(vision_tower, vision_tower_cfg=vision_tower_cfg, **kwargs)