Commit
·
127f135
1
Parent(s):
09bf719
Add a dataset loading script
Browse files- README.md +38 -0
- aio-passages.py +77 -0
README.md
ADDED
@@ -0,0 +1,38 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
language:
|
3 |
+
- ja
|
4 |
+
size_categories:
|
5 |
+
- 1M<n<10M
|
6 |
+
license:
|
7 |
+
- cc-by-sa-3.0
|
8 |
+
- gfdl
|
9 |
+
dataset_info:
|
10 |
+
features:
|
11 |
+
- name: id
|
12 |
+
dtype: int32
|
13 |
+
- name: pageid
|
14 |
+
dtype: int32
|
15 |
+
- name: revid
|
16 |
+
dtype: int32
|
17 |
+
- name: text
|
18 |
+
dtype: string
|
19 |
+
- name: section
|
20 |
+
dtype: string
|
21 |
+
- name: title
|
22 |
+
dtype: string
|
23 |
+
splits:
|
24 |
+
- name: train
|
25 |
+
num_bytes: 3054493919
|
26 |
+
num_examples: 4288198
|
27 |
+
download_size: 1110830651
|
28 |
+
dataset_size: 3054493919
|
29 |
+
---
|
30 |
+
|
31 |
+
# Dataset Card for llm-book/aio-passages
|
32 |
+
|
33 |
+
書籍『大規模言語モデル入門』で使用する、「AI王」コンペティションのパッセージデータセットです。
|
34 |
+
GitHub リポジトリ [cl-tohoku/quiz-datasets](https://github.com/cl-tohoku/quiz-datasets) で公開されているデータセットを利用しています。
|
35 |
+
|
36 |
+
## Licence
|
37 |
+
|
38 |
+
本データセットで利用している Wikipedia のコンテンツは、[クリエイティブ・コモンズ表示・継承ライセンス 3.0 (CC BY-SA 3.0)](https://creativecommons.org/licenses/by-sa/3.0/deed.ja) および [GNU 自由文書ライセンス (GFDL)](https://www.gnu.org/licenses/fdl.html) の下に配布されているものです。
|
aio-passages.py
ADDED
@@ -0,0 +1,77 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Copyright 2020 The HuggingFace Datasets Authors.
|
2 |
+
# Copyright 2023 Masatoshi Suzuki (@singletongue).
|
3 |
+
#
|
4 |
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
5 |
+
# you may not use this file except in compliance with the License.
|
6 |
+
# You may obtain a copy of the License at
|
7 |
+
#
|
8 |
+
# http://www.apache.org/licenses/LICENSE-2.0
|
9 |
+
#
|
10 |
+
# Unless required by applicable law or agreed to in writing, software
|
11 |
+
# distributed under the License is distributed on an "AS IS" BASIS,
|
12 |
+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
13 |
+
# See the License for the specific language governing permissions and
|
14 |
+
# limitations under the License.
|
15 |
+
import io
|
16 |
+
from typing import Iterator, List, Tuple
|
17 |
+
|
18 |
+
import datasets
|
19 |
+
import pyarrow as pa
|
20 |
+
|
21 |
+
|
22 |
+
_DESCRIPTION = (
|
23 |
+
"書籍『大規模言語モデル入門』で使用する、「AI王」コンペティションのパッセージデータセットです。"
|
24 |
+
"GitHub リポジトリ cl-tohoku/quiz-datasets で公開されているデータセットを利用しています。"
|
25 |
+
)
|
26 |
+
_HOMEPAGE = "https://github.com/cl-tohoku/quiz-datasets"
|
27 |
+
_LICENSE = (
|
28 |
+
"本データセットで使用している Wikipedia のコンテンツは、クリエイティブ・コモンズ表示・継承ライセンス 3.0 (CC BY-SA 3.0) "
|
29 |
+
"および GNU 自由文書ライセンス (GFDL) の下に配布されているものです。"
|
30 |
+
)
|
31 |
+
|
32 |
+
_URL_BASE = "https://github.com/cl-tohoku/quiz-datasets/releases/download"
|
33 |
+
_URL = f"{_URL_BASE}/v1.0.1/passages.jawiki-20220404-c400-large.jsonl.gz"
|
34 |
+
|
35 |
+
|
36 |
+
class AioPassages(datasets.ArrowBasedBuilder):
|
37 |
+
VERSION = datasets.Version("1.0.0")
|
38 |
+
|
39 |
+
def _info(self) -> datasets.DatasetInfo:
|
40 |
+
features = datasets.Features({
|
41 |
+
"id": datasets.Value("int32"),
|
42 |
+
"pageid": datasets.Value("int32"),
|
43 |
+
"revid": datasets.Value("int32"),
|
44 |
+
"text": datasets.Value("string"),
|
45 |
+
"section": datasets.Value("string"),
|
46 |
+
"title": datasets.Value("string"),
|
47 |
+
})
|
48 |
+
return datasets.DatasetInfo(
|
49 |
+
description=_DESCRIPTION,
|
50 |
+
homepage=_HOMEPAGE,
|
51 |
+
license=_LICENSE,
|
52 |
+
features=features,
|
53 |
+
)
|
54 |
+
|
55 |
+
def _split_generators(self, dl_manager: datasets.DownloadManager) -> List[datasets.SplitGenerator]:
|
56 |
+
filepath = dl_manager.download_and_extract(_URL)
|
57 |
+
|
58 |
+
split_generators = [datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": filepath})]
|
59 |
+
return split_generators
|
60 |
+
|
61 |
+
def _generate_tables(self, filepath: str, chunksize: int = 10 << 20) -> Iterator[Tuple[int, pa.Table]]:
|
62 |
+
# cf. https://github.com/huggingface/datasets/blob/2.12.0/src/datasets/packaged_modules/json/json.py
|
63 |
+
with open(filepath, "rb") as f:
|
64 |
+
batch_idx = 0
|
65 |
+
block_size = max(chunksize // 32, 16 << 10)
|
66 |
+
while True:
|
67 |
+
batch = f.read(chunksize)
|
68 |
+
if not batch:
|
69 |
+
break
|
70 |
+
|
71 |
+
batch += f.readline()
|
72 |
+
pa_table = pa.json.read_json(
|
73 |
+
io.BytesIO(batch), read_options=pa.json.ReadOptions(block_size=block_size)
|
74 |
+
)
|
75 |
+
|
76 |
+
yield batch_idx, pa_table
|
77 |
+
batch_idx += 1
|