preprocess.py

from transformers import T5Tokenizer
import polars as pl
from tqdm import tqdm
from functools import partial


def detokenize(seq: str, tokenizer: T5Tokenizer):
    output = tokenizer.decode(seq)
    output = output.replace(" ", "")
    output = output.replace("</s>", "")

    assert len(output) == len(seq[:-1])
    return output


def main():
    tokenizer = T5Tokenizer.from_pretrained('Rostlab/ProstT5', do_lower_case=False) # noqa
    splits = {'test': 'data/test-00000-of-00001-b109fa020c25190c.parquet', 'valid': 'data/valid-00000-of-00001-6442282fee0bc004.parquet', 'train': 'data/train-*-of-*.parquet'} # noqa

    detokenize_func = partial(detokenize, tokenizer=tokenizer)

    for k, v in tqdm(splits.items()):
        df = pl.scan_parquet('hf://datasets/Rostlab/ProstT5Dataset/' + v) # noqa
        df = df.with_columns(pl.col("input_id_x").map_elements(detokenize_func, return_dtype=pl.String).alias("3di")) # noqa
        df = df.with_columns(pl.col("input_id_y").map_elements(detokenize_func, return_dtype=pl.String).alias("protein")) # noqa
        df = df.drop("input_id_x").drop("input_id_y")
        df.sink_parquet(k + ".parquet")


if __name__ == "__main__":
    main()