LLaMA-Factory/data/ultra_chat/ultra_chat.py

61 lines
2.3 KiB
Python
Raw Normal View History

2023-05-28 10:09:04 +00:00
import json
2024-04-19 17:31:38 +00:00
import os
2023-11-09 07:53:23 +00:00
from typing import List
2023-05-28 10:09:04 +00:00
2024-04-19 17:31:38 +00:00
import datasets
2024-03-20 12:09:06 +00:00
_HF_ENDPOINT = os.getenv("HF_ENDPOINT", "https://huggingface.co")
2023-05-28 10:09:04 +00:00
_DESCRIPTION = "UltraChat: Large-scale, Informative, and Diverse Multi-round Dialogue Data."
_CITATION = """\
@misc{UltraChat,
author = {Ding, Ning and Chen, Yulin and Xu, Bokai and Hu, Shengding and Qin, Yujia and Liu, Zhiyuan and Sun, Maosong and Zhou, Bowen},
title = {UltraChat: A Large-scale Auto-generated Multi-round Dialogue Data},
year = {2023},
publisher = {GitHub},
journal = {GitHub repository},
howpublished = {\\url{https://github.com/thunlp/ultrachat}},
}
"""
2024-03-20 12:09:06 +00:00
_HOMEPAGE = "{}/datasets/stingning/ultrachat".format(_HF_ENDPOINT)
2023-05-28 10:09:04 +00:00
_LICENSE = "cc-by-nc-4.0"
2024-03-20 12:09:06 +00:00
_BASE_DATA_URL = "{}/datasets/stingning/ultrachat/resolve/main/train_{{idx}}.jsonl".format(_HF_ENDPOINT)
2023-05-28 10:09:04 +00:00
2023-11-09 07:53:23 +00:00
class UltraChat(datasets.GeneratorBasedBuilder):
2023-05-28 10:09:04 +00:00
VERSION = datasets.Version("0.0.0")
2023-11-09 07:53:23 +00:00
def _info(self):
2024-04-19 17:31:38 +00:00
features = datasets.Features(
{"conversations": [{"from": datasets.Value("string"), "value": datasets.Value("string")}]}
)
2023-05-28 10:09:04 +00:00
return datasets.DatasetInfo(
2024-04-19 17:31:38 +00:00
description=_DESCRIPTION, features=features, homepage=_HOMEPAGE, license=_LICENSE, citation=_CITATION
2023-05-28 10:09:04 +00:00
)
2023-11-09 07:53:23 +00:00
def _split_generators(self, dl_manager: datasets.DownloadManager):
2024-04-19 17:31:38 +00:00
file_paths = [dl_manager.download(_BASE_DATA_URL.format(idx=idx)) for idx in range(10)] # multiple shards
return [datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepaths": file_paths})]
2023-05-28 10:09:04 +00:00
2023-11-09 07:53:23 +00:00
def _generate_examples(self, filepaths: List[str]):
2023-05-28 10:09:04 +00:00
for filepath in filepaths:
with open(filepath, "r", encoding="utf-8") as f:
for row in f:
try:
data = json.loads(row)
2024-04-19 17:31:38 +00:00
except Exception:
2023-05-28 10:09:04 +00:00
continue
2023-11-09 07:53:23 +00:00
key: int = data["id"]
content: List[str] = data["data"]
2023-05-28 10:09:04 +00:00
if len(content) % 2 == 1:
content.pop(-1)
if len(content) < 2:
continue
2024-04-19 17:31:38 +00:00
conversations = [
{"from": "human" if i % 2 == 0 else "gpt", "value": content[i]} for i in range(len(content))
]
2023-11-15 18:08:04 +00:00
yield key, {"conversations": conversations}