convert_and_upload.py

import pandas as pd
import os
import datasets

XLS_FOLDER = "woori"
OUTPUT_FOLDER = "data"
SORT_KEY = "어휘"
NUM_PROC =32
hf_access_token = "YOUR_TOKEN"
hf_ID = "YOUR_ID"
ds_name = "woori_spring_dict"


def flatten_examples(example: dict) -> dict:
    text_line = ""
    for key in example:
        # some columns are empty or invalid
        if key == "의미 번호":
            continue
        if (single_column := example[key]) == None:
            continue
        # certain columns contain extraneous content
        if key == "원어·어종":
            single_column = single_column.removeprefix("안 밝힘 ")
        text_line += key + ": " + single_column.strip() + ", "
    return {"text": text_line.removesuffix(", ")}


if os.path.exists(XLS_FOLDER):
    XLS_FOLDER = os.path.abspath(XLS_FOLDER)
    xls_list = os.listdir(XLS_FOLDER)
else:
    raise ValueError("input folder does not exist")

combined_df = pd.DataFrame()
length_check = 0
for xls in sorted(xls_list):
    xls_path = os.path.join(XLS_FOLDER, xls)
    if os.path.exists(xls_path):
        # print(xls_path)
        df = pd.read_excel(xls_path, header=0, index_col=None)
        length_check += len(df)
        combined_df = pd.concat([combined_df, df], ignore_index=True)

assert len(combined_df) == length_check
ds = datasets.Dataset.from_pandas(combined_df)
sorted_ds = ds.sort(SORT_KEY)
processed_ds = sorted_ds.map(flatten_examples, num_proc=NUM_PROC).select_columns("text")

processed_ds.push_to_hub(repo_id=ds_name, token=hf_access_token)

# processed_ds.to_json("./woorimal.jsonl",num_proc=32,force_ascii=False,indent=1)