testbed/embeddings-benchmark__mteb/scripts/data/summeval_fr/create_data.py

from __future__ import annotations

import json
import os

import datasets
import requests
from huggingface_hub import create_repo, upload_file

# API key DeepL to set before running the script with the command 'export DEEPL_API_KEY=***'
DEEPL_API_KEY = os.environ.get("DEEPL_AUTH_KEY")
ORGANIZATION = "lyon-nlp/"
REPO_NAME = "summarization-summeval-fr-p2p"
SAVE_PATH = "test.json"
HEADERS = {
    "Authorization": f"DeepL-Auth-Key {DEEPL_API_KEY}",
    "Content-Type": "application/x-www-form-urlencoded",
}


def translate_with_deepl(text: str) -> str:
    data = {
        "text": text,
        "target_lang": "FR",
    }
    response = requests.post(
        "https://api.deepl.com/v2/translate", headers=HEADERS, data=data
    )
    return response.json()["translations"][0]["text"]


summeval = datasets.load_dataset("mteb/summeval")["test"]

trads = []

for line in summeval:
    trad = line
    trad["text"] = translate_with_deepl(line["text"])

    machine_summaries = []
    for machine_sum in line["machine_summaries"]:
        machine_summaries.append(translate_with_deepl(machine_sum))
    trad["machine_summaries"] = machine_summaries

    human_summaries = []
    for human_sum in line["human_summaries"]:
        human_summaries.append(translate_with_deepl(human_sum))
    trad["human_summaries"] = human_summaries
    trads.append(trad)

    with open(SAVE_PATH, "w", encoding="utf8") as final:
        json.dump(trads, final, ensure_ascii=False)


create_repo(ORGANIZATION + REPO_NAME, repo_type="dataset")

upload_file(
    path_or_fileobj=SAVE_PATH,
    path_in_repo=SAVE_PATH,
    repo_id=ORGANIZATION + REPO_NAME,
    repo_type="dataset",
)
os.system(f"rm {SAVE_PATH}")