data_prep/original_v1/extract.py

import os
import requests
from glob import glob
import time

from datasets import Dataset

langs = [
    "asm",
    "awa",
    # "bal",
    "ben",
    "bho",
    # "bod",
    "brx",
    # "div",
    "guj",
    # "hif",
    "hin",
    "kan",
    "kha",
    "kok",
    # "lah",
    "mai",
    "mal",
    "mar",
    "mni",
    "nep",
    # "nst",
    "ori",
    "pan",
    "pli",
    # "rhg",
    "san",
    "sat",
    # "sin",
    "snd",
    "tam",
    "tel",
    "urd",
]

HF_REPO = "sarvamai/tatoeba-indic"
BASE_URL = "https://object.pouta.csc.fi/Tatoeba-Challenge-v2023-09-26/{direction}.tar"

for lang in langs:
    direction = f"eng-{lang}"
    url1 = BASE_URL.format(direction=direction)
    # Check if the file exists
    response = requests.head(url1)
    if response.status_code == 200:
        print(f"File exists: {url1}")
        download_url = url1
    else:
        direction = f"{lang}-eng"
        url2 = BASE_URL.format(direction=direction)
        response = requests.head(url2)
        if response.status_code == 200:
            print(f"File exists: {url2}")
            download_url = url2
        else:
            raise ValueError(f"File does not exist: {url1} or {url2}")
    
    file_name = download_url.split("/")[-1]
    if not os.path.exists(file_name):
        # Download the file
        os.system(f"wget {download_url}")

        # Extract the file
        os.system(f"tar -xvf {file_name}")

    src_lang, tgt_lang = direction.split("-")

    sets = ["test", "dev"]

    for split in sets:
        src_files = glob(f"data/release/**/{direction}/{split}.src")
        if not src_files:
            print(f"No {split} src files found for {direction}")
            continue

        tgt_files = glob(f"data/release/**/{direction}/{split}.trg")

        src_lines, tgt_lines = [], []
        for src_line in open(src_files[0]):
            src_lines.append(src_line.strip())

        for tgt_line in open(tgt_files[0]):
            tgt_lines.append(tgt_line.strip())

        assert len(src_lines) == len(tgt_lines), f"Lengths of {split} src and tgt files are not equal for {direction}"

        if src_lang != "eng":
            # Swap src and tgt
            src_lines, tgt_lines = tgt_lines, src_lines

        dataset = Dataset.from_dict({"src": src_lines, "tgt": tgt_lines})
        dataset.push_to_hub(HF_REPO, lang, split=split, commit_message=f"Add {split} split for {lang} language"); time.sleep(2)