docs: enhance comment in config/train_cord.yaml

config/train_cord.yaml

@@ -5,7 +5,7 @@ dataset_name_or_paths: ["naver-clova-ix/cord-v2"] # loading datasets (from molde
 sort_json_key: False # cord dataset is preprocessed, and publicly available at https://huggingface.co/datasets/naver-clova-ix/cord-v2
 train_batch_sizes: [8]
 val_batch_sizes: [1]
-input_size: [1280, 960]
+input_size: [1280, 960] # when the input resolution differs from the pre-training setting, some weights will be newly initialized (but the model training would be okay)
 max_length: 768
 align_long_axis: False
 num_nodes: 1