train_file: ['data/coco.json',
               'data/vg.json',
               'data/cc3m_train.json',
               'data/cc3m_val.json',
               'data/sbu.json'               
               ]

# each train_file (json) contains a python list where each item is {'image': img_path, 'caption': text or list_of_text }  

bert_config: 'configs/config_bert.json'

image_res: 256
vision_width: 768
embed_dim: 256
batch_size: 64

optimizer: {opt: adamW, lr: 2e-5, weight_decay: 0.02}
schedular: {sched: cosine, lr: 2e-5, epochs: 1, min_lr: 1e-5, decay_rate: 1, warmup_lr: 1e-5, warmup_epochs: 1, cooldown_epochs: 0}