def full(tokenizer: PreTrainedTokenizer, transform: object = None, seq_len: int = 512, tgt_seq_len: int = 32, no_image: bool = False): dataset = ConcatDataset([DocVQA(mode, tokenizer, transform=transform, seq_len=seq_len, no_image=no_image, tgt_seq_len=tgt_seq_len) for mode in ["train", "val", "test"]]) dataset.__setattr__("tokenizer", tokenizer) return dataset