# -*- encoding: utf-8 -*-
'''
@Time    :   2023-06-19 15:50:00
@Author  :   宋昊阳
@Contact :   1627635056@qq.com
'''
import json
from tqdm import tqdm
from datasets import load_dataset


raw_data = [json.loads(i) for i in tqdm(open("./HuatuoGPT_sft_data_v1.jsonl", 'r', encoding='utf-8').readlines())]
def clean_str(string):
    return string.replace("问：", "").replace("答：", "").strip()
data = []
for item in tqdm(raw_data):
    metadata = {"instruction": "","input": "","output": "","history": []}
    idx = 0
    for seq in item['data'][:-2]:
        seq = clean_str(seq)
        metadata['history'].append(seq)
    metadata['history'] = [metadata['history'][2*i:2*(i+1)] for i in range(len(metadata['history'])//2)] if metadata['history'] else []
    metadata["instruction"] = clean_str(item['data'][-2])
    metadata["output"] = clean_str(item['data'][-1])
    data.append(metadata)
json.dump(data, open("HuatuoGPT_sft_data_v1.json", "w", encoding="utf-8"), ensure_ascii=False, indent=4)
data = load_dataset("json", data_files="HuatuoGPT_sft_data_v1.json")
print(data.column_names['train'], data.num_rows)