daichira commited on
Commit
fd733d7
·
verified ·
1 Parent(s): 88f0565

Upload 3 files

Browse files
Files changed (3) hide show
  1. ATTRIBUTION.md +16 -0
  2. LICENSE.md +22 -0
  3. README.md +148 -25
ATTRIBUTION.md ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Attribution & Sources
2
+
3
+ This dataset thanks the following sources:
4
+
5
+ - OpenFoodFacts product-database
6
+ - ODbL/DbCL for database and contents; images CC BY-SA (not included here)
7
+ - https://world.openfoodfacts.org/data
8
+ - Shopify/product-catalogue
9
+ - Dataset card / terms:
10
+ - https://huggingface.co/datasets/Shopify/product-catalogue
11
+ - ontologicalapple/vrts-gtfs-archive
12
+ - Open data terms (e.g., BC Transit Open Data Terms of Use) apply
13
+ - https://huggingface.co/datasets/ontologicalapple/vrts-gtfs-archive
14
+
15
+ If you create derivative works from this dataset, please include attribution language referencing the above sources and licenses in your own documentation and model cards.
16
+
LICENSE.md ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # License / Terms
2
+
3
+ This dataset (“AppliedCourse_SFT_datasets_v5”) aggregates and transforms content sourced from third-party open datasets. It is distributed under “other” to reflect that upstream licenses and terms apply. You must comply with all applicable upstream licenses and terms when using this dataset and any derivatives.
4
+
5
+ Important notes:
6
+
7
+ - OpenFoodFacts product-database
8
+ - Database: Open Database License (ODbL 1.0)
9
+ - Contents: Database Contents License (DbCL 1.0)
10
+ - Images: CC BY-SA (images are not redistributed in this dataset)
11
+ - See: https://world.openfoodfacts.org/data
12
+ - Shopify/product-catalogue
13
+ - See the dataset card and license/terms on Hugging Face:
14
+ - https://huggingface.co/datasets/Shopify/product-catalogue
15
+ - ontologicalapple/vrts-gtfs-archive
16
+ - Governed by the open data terms (e.g., BC Transit Open Data Terms of Use) linked from the dataset card:
17
+ - https://huggingface.co/datasets/ontologicalapple/vrts-gtfs-archive
18
+
19
+ No ownership of upstream content is claimed. This repository does not redistribute upstream datasets. The dataset only contains derived text rows and prompts needed for supervised fine-tuning.
20
+
21
+ Redistribution, public sharing, and commercial use may require further obligations (e.g., attribution) depending on upstream licenses. It is your responsibility to review and comply with all requirements in the relevant jurisdictions.
22
+
README.md CHANGED
@@ -1,31 +1,154 @@
1
  ---
2
- dataset_info:
3
- features:
4
- - name: id
5
- dtype: string
6
- - name: category
7
- dtype: string
8
- - name: subcategory
9
- dtype: string
10
- - name: task
11
- dtype: string
12
- - name: seed
13
- dtype: string
14
- - name: messages
15
- list:
16
- - name: role
17
- dtype: string
18
- - name: content
19
- dtype: string
20
- splits:
21
- - name: train
22
- num_bytes: 15199277
23
- num_examples: 18800
24
- download_size: 2791255
25
- dataset_size: 15199277
26
  configs:
27
  - config_name: default
28
  data_files:
29
  - split: train
30
- path: data/train-*
31
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ language:
3
+ - ja
4
+ - en
5
+ license: other
6
+ tags:
7
+ - sft
8
+ - jsonl
9
+ - openai-messages
10
+ - structured-output
11
+ - format-conversion
12
+ - information-extraction
13
+ - csv
14
+ - json
15
+ - yaml
16
+ - xml
17
+ - toml
18
+ dataset_type: jsonl
19
+ pretty_name: AppliedCourse_SFT_datasets_v5
20
+ size_categories:
21
+ - 10K<n<50K
 
 
 
 
22
  configs:
23
  - config_name: default
24
  data_files:
25
  - split: train
26
+ path: "*.jsonl"
27
  ---
28
+
29
+ # AppliedCourse_SFT_datasets_v5
30
+
31
+ 本データセットは、構造化出力(CSV / JSON / XML / TOML / YAML)の相互変換および抽出タスク用の SFT 学習データです。各サンプルは OpenAI `messages` 形式(`[{role, content}, ...]`)のJSONLで提供され、assistant の出力は仕様に従うよう設計されています。
32
+
33
+ 重要: 本データセットは複数の公開データソースに依拠しており、利用時は必ず下記「Sources & Terms」に記載の各ライセンス・利用規約に従ってください。
34
+
35
+ ## Dataset Summary
36
+
37
+ - 目的: 構造化出力の形式変換(JSON↔YAML、JSON→CSV等)と情報抽出(CSV/XML/Text→JSONなど)を高品質に学習させる
38
+ - フォーマット: OpenAI `messages` 形式の JSONL
39
+ - 含まれる主なタスク例
40
+ - JSON↔CSV / JSON↔YAML / YAML→CSV / XML→CSV
41
+ - XML→JSON / Text→JSON / Text→YAML / Text→TOML
42
+ - JSON→XML / YAML→XML / CSV→XML / XML→YAML / TOML→JSON / YAML→TOML
43
+ - 品質設計(生成時)
44
+ - 構文検証(XML/TOML/YAML/CSV)を通過したサンプルのみ採用
45
+ - P0ガード(監督範囲ゼロを拒否)
46
+ - 重複除外、サイズ制御(行/属性/セル長)
47
+
48
+ ## Supported Tasks
49
+
50
+ - text2json / text2yaml / text2toml
51
+ - format-conversion: json↔yaml, json→csv, yaml→csv, xml→csv, json→xml, yaml→xml, csv→xml, xml→yaml, toml→json, yaml→toml
52
+ - information-extraction: csv→json, xml→json, text→json
53
+
54
+ ## Languages
55
+ - 日本語 / 英語(タスク指示は英語が中心)
56
+
57
+ ## Dataset Structure
58
+
59
+ - ファイル: `*.jsonl`(train split 想定)
60
+ - 1行 = 1サンプル(OpenAI messages形式)
61
+ - 推奨カラム(JSONオブジェクトのトップレベル)
62
+ - `id`: string(安定ハッシュ)
63
+ - `category`: string(内部カテゴリ)
64
+ - `subcategory`: string(タスク名、例: `json_to_xml`)
65
+ - `task`: string(extract / transform / filter)
66
+ - `seed`: string(元データの組み合わせ/出所ヒント)
67
+ - `messages`: list of `{role: "user"|"assistant", content: str}`
68
+
69
+ ### Example
70
+
71
+ ```json
72
+ {
73
+ "id": "a1b2c3d4e5f6",
74
+ "category": "C_XML",
75
+ "subcategory": "json_to_xml",
76
+ "task": "transform",
77
+ "seed": "openfoodfacts:default",
78
+ "messages": [
79
+ {"role": "user", "content": "Convert the following JSON into well-formed XML...\nJSON:\n{\"items\": [{\"name\": \"A\"}, {\"name\": \"B\"}]}\n"},
80
+ {"role": "assistant", "content": "<root><items><item><name>A</name></item><item><name>B</name></item></items></root>"}
81
+ ]
82
+ }
83
+ ```
84
+
85
+ ## Data Splits
86
+
87
+ - `train` のみ(評価に使う際はユーザー側で分割してください)
88
+
89
+ ## Data Fields
90
+
91
+ - `messages[0].content` に入力の仕様・ルール・サンプルデータが含まれます
92
+ - `messages[-1].content` はフォーマット仕様に厳密に従った出力(ONLY <FORMAT>)
93
+
94
+ ## Collection Process
95
+
96
+ - Hugging Face上の公開データセットから安全な列をストリーミング取得し、小規模な正規化を行った上でタスク化
97
+ - 各フォーマットの生成では、直列化ルールをプロンプトで明示
98
+ - 検証(構文/往復/属性)とP0ガードを適用
99
+ - 重複削除、形式分布のチューニングも実施
100
+
101
+ ## Sources & Terms(IMPORTANT)
102
+
103
+ 本データセットは下記の公開データに依拠しています。利用にあたっては各ライセンス・利用条件を遵守してください。
104
+
105
+ - OpenFoodFacts product-database
106
+ - Database: Open Database License (ODbL 1.0)
107
+ - Contents: Database Contents License (DbCL 1.0)
108
+ - Images: CC BY-SA(本データセットには画像は含みません)
109
+ - 参考: https://world.openfoodfacts.org/data
110
+ - Shopify/product-catalogue
111
+ - データセットのカードと利用条件を参照
112
+ - 参考: https://huggingface.co/datasets/Shopify/product-catalogue
113
+ - ontologicalapple/vrts-gtfs-archive
114
+ - 掲載の Open Data Terms(例: BC Transit Open Data Terms of Use)に準拠
115
+ - 参考: https://huggingface.co/datasets/ontologicalapple/vrts-gtfs-archive
116
+
117
+ 注意: 本リポジトリは上記のデータを**再配布しません**。��生データ(タスク化・整形)であっても、元ライセンスの義務(帰属等)は引き続きユーザーに帰属します。
118
+
119
+ ## Licensing Information
120
+
121
+ - ライセンス: `other`(上記の upstream ライセンス/規約を順守)
122
+ - 本データセットの利用に伴い、元データのライセンス条件(ODbL/DbCL/CC BY-SA 等)に従う必要があります
123
+ - 本リポジトリのファイル構成や付属のTERMS/ATTRIBUTIONは、順守事項の理解を補助するためのものです
124
+
125
+ ## Intended Use
126
+
127
+ - 構造化出力タスクのSFT(CSV/JSON/XML/TOML/YAMLの変換・抽出)
128
+ - 研究・教育目的のベースライン構築
129
+
130
+ ## Limitations
131
+
132
+ - 仕様が曖昧な指示に対する出力は保証されません
133
+ - 多言語・特殊文字・空値混在などの難ケースが含まれます
134
+
135
+ ## How to Use
136
+
137
+ ```python
138
+ from datasets import load_dataset
139
+
140
+ ds = load_dataset("daichira/AppliedCourse_SFT_datasets_v5", split="train")
141
+ print(len(ds), ds.column_names)
142
+ print(ds[0]["messages"][0]["content"])
143
+ ```
144
+
145
+ ## Citation
146
+
147
+ - OpenFoodFacts: https://world.openfoodfacts.org/
148
+ - Shopify/product-catalogue: https://huggingface.co/datasets/Shopify/product-catalogue
149
+ - VRTS GTFS archive: https://huggingface.co/datasets/ontologicalapple/vrts-gtfs-archive
150
+
151
+ ## Contributions
152
+
153
+ Issues / PRs welcome. 上流ライセンス遵守と本データカードの更新にご協力ください。
154
+