Update files from the datasets library (from 1.8.0)

Release notes: https://github.com/huggingface/datasets/releases/tag/1.8.0

.gitattributes

@@ -0,0 +1,27 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bin.* filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zstandard filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,441 @@
+---
+annotations_creators:
+- expert-generated
+language_creators:
+- expert-generated
+languages:
+- ko
+licenses:
+- cc-by-sa-4-0
+multilinguality:
+- monolingual
+size_categories:
+- 10K<n<100K
+source_datasets:
+- original
+task_categories:
+  ynat:
+  - text-classification
+  sts:
+  - text-scoring
+  nli:
+  - text-classification
+  ner:
+  - structure-prediction
+  re:
+  - structure-prediction
+  dp:
+  - structure-prediction
+  mrc:
+  - question-answering
+  wos:
+  - sequence-modeling
+task_ids:
+  ynat:
+  - topic-classification
+  sts:
+  - semantic-similarity-scoring
+  nli:
+  - natural-language-inference
+  ner:
+  - named-entity-recognition
+  re:
+  - other-relation-extraction
+  dp:
+  - parsing
+  mrc:
+  - extractive-qa
+  wos:
+  - other-dialogue-state-tracking
+paperswithcode_id: klue
+---
+
+# Dataset Card for KLUE
+
+## Table of Contents
+- [Dataset Description](#dataset-description)
+  - [Dataset Summary](#dataset-summary)
+  - [Supported Tasks](#supported-tasks-and-leaderboards)
+  - [Languages](#languages)
+- [Dataset Structure](#dataset-structure)
+  - [Data Instances](#data-instances)
+  - [Data Fields](#data-instances)
+  - [Data Splits](#data-instances)
+- [Dataset Creation](#dataset-creation)
+  - [Curation Rationale](#curation-rationale)
+  - [Source Data](#source-data)
+  - [Annotations](#annotations)
+  - [Personal and Sensitive Information](#personal-and-sensitive-information)
+- [Considerations for Using the Data](#considerations-for-using-the-data)
+  - [Social Impact of Dataset](#social-impact-of-dataset)
+  - [Discussion of Biases](#discussion-of-biases)
+  - [Other Known Limitations](#other-known-limitations)
+- [Additional Information](#additional-information)
+  - [Dataset Curators](#dataset-curators)
+  - [Licensing Information](#licensing-information)
+  - [Citation Information](#citation-information)
+
+## Dataset Description
+
+- **Homepage:** https://klue-benchmark.com/
+- **Repository:** https://github.com/KLUE-benchmark/KLUE
+- **Paper:** [KLUE: Korean Language Understanding Evaluation](https://arxiv.org/abs/2105.09680)
+- **Leaderboard:** [Leaderboard](https://klue-benchmark.com/leaderboard)
+- **Point of Contact:** https://github.com/KLUE-benchmark/KLUE/issues
+
+### Dataset Summary
+
+KLUE is a collection of 8 tasks to evaluate natural language understanding capability of Korean language models. We delibrately select the 8 tasks, which are Topic Classification, Semantic Textual Similarity, Natural Language Inference, Named Entity Recognition, Relation Extraction, Dependency Parsing, Machine Reading Comprehension, and Dialogue State Tracking.
+
+### Supported Tasks and Leaderboards
+
+Topic Classification, Semantic Textual Similarity, Natural Language Inference, Named Entity Recognition, Relation Extraction, Dependency Parsing, Machine Reading Comprehension, and Dialogue State Tracking
+
+### Languages
+
+`ko-KR`
+
+## Dataset Structure
+
+### Data Instances
+
+#### ynat
+An example of 'train' looks as follows.
+
+```
+{'date': '2016.06.30. 오전 10:36',
+ 'guid': 'ynat-v1_train_00000',
+ 'label': 3,
+ 'title': '유튜브 내달 2일까지 크리에이터 지원 공간 운영',
+ 'url': 'https://news.naver.com/main/read.nhn?mode=LS2D&mid=shm&sid1=105&sid2=227&oid=001&aid=0008508947'}
+```
+
+#### sts
+An example of 'train' looks as follows.
+
+```
+{'guid': 'klue-sts-v1_train_00000',
+ 'labels': {'label': 3.7, 'real-label': 3.714285714285714, 'binary-label': 1},
+ 'sentence1': '숙소 위치는 찾기 쉽고 일반적인 한국의 반지하 숙소입니다.',
+ 'sentence2': '숙박시설의 위치는 쉽게 찾을 수 있고 한국의 대표적인 반지하 숙박시설입니다.',
+ 'source': 'airbnb-rtt'}
+```
+
+#### nli
+An example of 'train' looks as follows.
+
+```
+{'guid': 'klue-nli-v1_train_00000',
+ 'hypothesis': '힛걸 진심 최고로 멋지다.',
+ 'label': 0,
+ 'premise': '힛걸 진심 최고다 그 어떤 히어로보다 멋지다',
+ 'source': 'NSMC'}
+```
+
+#### ner
+An example of 'train' looks as follows.
+
+```
+{'tokens': ['특', '히', ' ', '영', '동', '고', '속', '도', '로', ' ', '강', '릉', ' ', '방', '향', ' ', '문', '막', '휴', '게', '소', '에', '서', ' ', '만', '종', '분', '기', '점', '까', '지', ' ', '5', '㎞', ' ', '구', '간', '에', '는', ' ', '승', '용', '차', ' ', '전', '용', ' ', '임', '시', ' ', '갓', '길', '차', '로', '제', '를', ' ', '운', '영', '하', '기', '로', ' ', '했', '다', '.'],
+ 'ner_tags': [12, 12, 12, 2, 3, 3, 3, 3, 3, 12, 2, 3, 12, 12, 12, 12, 2, 3, 3, 3, 3, 12, 12, 12, 2, 3, 3, 3, 3, 12, 12, 12, 8, 9, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12],
+ 'sentence': '특히 <영동고속도로:LC> <강릉:LC> 방향 <문막휴게소:LC>에서 <만종분기점:LC>까지 <5㎞:QT> 구간에는 승용차 전용 임시 갓길차로제를 운영하기로 했다.'}
+```
+
+#### re
+An example of 'train' looks as follows.
+
+```
+{'guid': 'klue-re-v1_train_00000',
+ 'label': 0,
+ 'object_entity': {'word': '조지 해리슨',
+  'start_idx': 13,
+  'end_idx': 18,
+  'type': 'PER'},
+ 'sentence': '〈Something〉는 조지 해리슨이 쓰고 비틀즈가 1969년 앨범 《Abbey Road》에 담은 노래다.',
+ 'source': 'wikipedia',
+ 'subject_entity': {'word': '비틀즈',
+  'start_idx': 24,
+  'end_idx': 26,
+  'type': 'ORG'}}
+```
+
+#### dp
+An example of 'train' looks as follows.
+
+```
+{'deprel': ['NP', 'NP_OBJ', 'VP', 'NP', 'NP_SBJ', 'NP', 'NP_MOD', 'NP_CNJ', 'NP_CNJ', 'NP', 'NP', 'NP_OBJ', 'AP', 'VP'],
+ 'head': [2, 3, 14, 5, 14, 7, 10, 10, 10, 11, 12, 14, 14, 0],
+ 'index': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14],
+ 'lemma': ['해당', '그림 을', '보 면', '디즈니', '공주 들 이', '브리트니', '스피어스 의', '앨범 이나', '뮤직 비디오 ,', '화보', '속', '모습 을', '똑같이', '재연 하 였 다 .'],
+ 'pos': ['NNG', 'NNG+JKO', 'VV+EC', 'NNP', 'NNG+XSN+JKS', 'NNP', 'NNP+JKG', 'NNG+JC', 'NNG+NNG+SP', 'NNG', 'NNG', 'NNG+JKO', 'MAG', 'NNG+XSA+EP+EF+SF'],
+ 'sentence': '해당 그림을 보면 디즈니 공주들이 브리트니 스피어스의 앨범이나 뮤직비디오, 화보 속 모습을 똑같이 재연했다.',
+ 'word_form': ['해당', '그림을', '보면', '디즈니', '공주들이', '브리트니', '스피어스의', '앨범이나', '뮤직비디오,', '화보', '속', '모습을', '똑같이', '재연했다.']}
+```
+
+#### mrc
+An example of 'train' looks as follows.
+
+```
+{'answers': {'answer_start': [478, 478], 'text': ['한 달가량', '한 달']},
+ 'context': '올여름 장마가 17일 제주도에서 시작됐다. 서울 등 중부지방은 예년보다 사나흘 정도 늦은 이달 말께 장마가 시작될 전망이다.17일 기상청에 따르면 제주도 남쪽 먼바다에 있는 장마전선의 영향으로 이날 제주도 산간 및 내륙지역에 호우주의보가 내려지면서 곳곳에 100㎜에 육박하는 많은 비가 내렸다. 제주의 장마는 평년보다 2~3일, 지난해보다는 하루 일찍 시작됐다. 장마는 고온다습한 북태평양 기단과 한랭 습윤한 오호츠크해 기단이 만나 형성되는 장마전선에서 내리는 비를 뜻한다.장마전선은 18일 제주도 먼 남쪽 해상으로 내려갔다가 20일께 다시 북상해 전남 남해안까지 영향을 줄 것으로 보인다. 이에 따라 20~21일 남부지방에도 예년보다 사흘 정도 장마가 일찍 찾아올 전망이다. 그러나 장마전선을 밀어올리는 북태평양 고기압 세력이 약해 서울 등 중부지방은 평년보다 사나흘가량 늦은 이달 말부터 장마가 시작될 것이라는 게 기상청의 설명이다. 장마전선은 이후 한 달가량 한반도 중남부를 오르내리며 곳곳에 비를 뿌릴 전망이다. 최근 30년간 평균치에 따르면 중부지방의 장마 시작일은 6월24~25일이었으며 장마기간은 32일, 강수일수는 17.2일이었다.기상청은 올해 장마기간의 평균 강수량이 350~400㎜로 평년과 비슷하거나 적을 것으로 내다봤다. 브라질 월드컵 한국과 러시아의 경기가 열리는 18일 오전 서울은 대체로 구름이 많이 끼지만 비는 오지 않을 것으로 예상돼 거리 응원에는 지장이 없을 전망이다.',
+ 'guid': 'klue-mrc-v1_train_12759',
+ 'is_impossible': False,
+ 'news_category': '종합',
+ 'question': '북태평양 기단과 오호츠크해 기단이 만나 국내에 머무르는 기간은?',
+ 'question_type': 1,
+ 'source': 'hankyung',
+ 'title': '제주도 장마 시작 … 중부는 이달 말부터'}
+```
+
+#### wos
+An example of 'train' looks as follows.
+
+```
+{'dialogue': [{'role': 'user',
+   'text': '쇼핑을 하려는데 서울 서쪽에 있을까요?',
+   'state': ['관광-종류-쇼핑', '관광-지역-서울 서쪽']},
+  {'role': 'sys',
+   'text': '서울 서쪽에 쇼핑이 가능한 곳이라면 노량진 수산물 도매시장이 있습니다.',
+   'state': []},
+  {'role': 'user',
+   'text': '오 네 거기 주소 좀 알려주세요.',
+   'state': ['관광-종류-쇼핑', '관광-지역-서울 서쪽', '관광-이름-노량진 수산물 도매시장']},
+  {'role': 'sys', 'text': '노량진 수산물 도매시장의 주소는 서울 동작구 93806입니다.', 'state': []},
+  {'role': 'user',
+   'text': '알려주시는김에 연락처랑 평점도 좀 알려주세요.',
+   'state': ['관광-종류-쇼핑', '관광-지역-서울 서쪽', '관광-이름-노량진 수산물 도매시장']},
+  {'role': 'sys', 'text': '그럼. 연락처는 6182006591이고 평점은 4점입니다.', 'state': []},
+  {'role': 'user',
+   'text': '와 감사합니다.',
+   'state': ['관광-종류-쇼핑', '관광-지역-서울 서쪽', '관광-이름-노량진 수산물 도매시장']},
+  {'role': 'sys', 'text': '감사합니다.', 'state': []}],
+ 'domains': ['관광'],
+ 'guid': 'wos-v1_train_00001'}
+```
+
+### Data Fields
+
+#### ynat
+
++ `guid`: a `string` feature
++ `title`: a `string` feature
++ `label`: a classification label, with possible values `IT과학`(0), `경제`(1), `사회`(2), `생활문화`(3), `세계`(4), `스포츠`(5), `정치`(6)
++ `url`: a `string` feature
++ `date`: a `string` feature
+
+#### sts
+
++ `guid`: a `string` feature
++ `source`: a `string` feature
++ `sentence1`: a `string` feature
++ `sentence2`: a `string` feature
++ `labels`: a dictionary feature containing
+    + `label`: a `float64` feature
+    + `real-label`: a `float64` feature
+    + `binary-label`: a classification label, with possible values `negative`(0), `positive`(1)
+  
+#### nli
+
++ `guid`: a `string` feature
++ `source`: a `string` feature
++ `premise`: a `string` feature
++ `hypothesis`: a `string` feature
++ `label`: a classification label, with possible values `entailment`(0), `neutral`(1), `contradiction`(2)
+
+#### ner
+
++ `sentence`: a `string` feature
++ `tokens`: a list of a `string` feature (tokenization is at character level)
++ `ner_tags`: a list of classification labels, with possible values including `B-DT`(0), `I-DT`(1), 
+  `B-LC`(2), `I-LC`(3), `B-OG`(4), `I-OG`(5), `B-PS`(6), `I-PS`(7), `B-QT`(8), `I-QT`(9), `B-TI`(10), 
+  `I-TI`(11), `O`(12)
+  
+#### re
+
++ `guid`: a `string` feature
++ `sentence`: a `string` feature
++ `subject_entity`: a dictionary feature containing
+    + `word`: a `string` feature
+    + `start_idx`: a `int32` feature 
+    + `end_idx`: a `int32` feature
+    + `type`: a `string` feature
++ `object_entity`: a dictionary feature containing
+    + `word`: a `string` feature
+    + `start_idx`: a `int32` feature 
+    + `end_idx`: a `int32` feature
+    + `type`: a `string` feature
++ `label`: a list of labels, with possible values including `no_relation`(0), `org:dissolved`(1), 
+  `org:founded`(2), `org:place_of_headquarters`(3), `org:alternate_names`(4), `org:member_of`(5), 
+  `org:members`(6), `org:political/religious_affiliation`(7), `org:product`(8), `org:founded_by`(9),`org:top_members/employees`(10), 
+  `org:number_of_employees/members`(11), `per:date_of_birth`(12), `per:date_of_death`(13), `per:place_of_birth`(14), 
+  `per:place_of_death`(15), `per:place_of_residence`(16), `per:origin`(17), `per:employee_of`(18), 
+  `per:schools_attended`(19), `per:alternate_names`(20), `per:parents`(21), `per:children`(22), 
+  `per:siblings`(23), `per:spouse`(24), `per:other_family`(25), `per:colleagues`(26), `per:product`(27), 
+  `per:religion`(28), `per:title`(29),
++ `source`: a `string` feature
+
+#### dp
+
++ `sentence`: a `string` feature
++ `index`: a list of `int32` feature 
++ `word_form`: a list of `string` feature
++ `lemma`: a list of `string` feature
++ `pos`: a list of `string` feature
++ `head`: a list of `int32` feature
++ `deprel`: a list of `string` feature
+
+
+#### mrc
+
++ `title`: a `string` feature
++ `context`: a `string` feature
++ `news_category`: a `string` feature
++ `source`: a `string` feature
++ `guid`: a `string` feature
++ `is_impossible`: a `bool` feature
++ `question_type`: a `int32` feature
++ `question`: a `string` feature
++ `answers`: a dictionary feature containing
+   + `answer_start`: a `int32` feature
+   + `text`: a `string` feature
+  
+
+#### wos
+
++ `guid`: a `string` feature
++ `domains`: a `string` feature
++ `dialogue`: a list of dictionary feature containing
+  + `role`: a `string` feature
+  + `text`: a `string` feature
+  + `state`: a `string` feature
+
+
+### Data Splits
+
+#### ynat
+
+You can see more details in [here](https://klue-benchmark.com/tasks/66/data/description).
+
++ train: 45,678
++ validation: 9,107
+
+
+#### sts
+
+You can see more details in [here](https://klue-benchmark.com/tasks/67/data/description).
+
++ train: 11,668
++ validation: 519
+
+#### nli
+
+You can see more details in [here](https://klue-benchmark.com/tasks/68/data/description).
+
++ train: 24,998
++ validation: 3,000
+
+#### ner
+
+You can see more details in [here](https://klue-benchmark.com/tasks/69/overview/description).
+
++ train: 21,008
++ validation: 5,000
+
+#### re
+
+You can see more details in [here](https://klue-benchmark.com/tasks/70/overview/description).
+
++ train: 32,470
++ validation: 7,765
+
+#### dp
+
+You can see more details in [here](https://klue-benchmark.com/tasks/71/data/description).
+
++ train: 10,000
++ validation: 2,000
+
+#### mrc
+
+You can see more details in [here](https://klue-benchmark.com/tasks/72/overview/description).
+
++ train: 17,554
++ validation: 5,841
+
+#### wos
+
+You can see more details in [here](https://klue-benchmark.com/tasks/73/overview/description).
+
++ train: 8,000
++ validation: 1,000
+
+
+## Dataset Creation
+
+### Curation Rationale
+
+[Needs More Information]
+
+### Source Data
+
+#### Initial Data Collection and Normalization
+
+[Needs More Information]
+
+#### Who are the source language producers?
+
+[Needs More Information]
+
+### Annotations
+
+#### Annotation process
+
+[Needs More Information]
+
+#### Who are the annotators?
+
+[Needs More Information]
+
+### Personal and Sensitive Information
+
+[Needs More Information]
+
+## Considerations for Using the Data
+
+### Social Impact of Dataset
+
+[Needs More Information]
+
+### Discussion of Biases
+
+[Needs More Information]
+
+### Other Known Limitations
+
+[Needs More Information]
+
+## Additional Information
+
+### Dataset Curators
+
+[Needs More Information]
+
+### Licensing Information
+
+[Needs More Information]
+
+### Citation Information
+
+```
+@misc{park2021klue,
+      title={KLUE: Korean Language Understanding Evaluation}, 
+      author={Sungjoon Park and Jihyung Moon and Sungdong Kim and Won Ik Cho and Jiyoon Han and Jangwon Park and Chisung Song and Junseong Kim and Yongsook Song and Taehwan Oh and Joohong Lee and Juhyun Oh and Sungwon Lyu and Younghoon Jeong and Inkwon Lee and Sangwoo Seo and Dongjun Lee and Hyunwoo Kim and Myeonghwa Lee and Seongbo Jang and Seungwon Do and Sunkyoung Kim and Kyungtae Lim and Jongwon Lee and Kyumin Park and Jamin Shin and Seonghyun Kim and Lucy Park and Alice Oh and Jungwoo Ha and Kyunghyun Cho},
+      year={2021},
+      eprint={2105.09680},
+      archivePrefix={arXiv},
+      primaryClass={cs.CL}
+}
+```
+### Contributions
+
+Thanks to [@jungwhank](https://github.com/jungwhank), [@bzantium](https://github.com/bzantium) for adding this dataset.

dataset_infos.json

@@ -0,0 +1 @@
+{"ynat": {"description": "KLUE (Korean Language Understanding Evaluation)\nKorean Language Understanding Evaluation (KLUE) benchmark is a series of datasets to evaluate natural language\nunderstanding capability of Korean language models. KLUE consists of 8 diverse and representative tasks, which are accessible\nto anyone without any restrictions. With ethical considerations in mind, we deliberately design annotation guidelines to obtain\nunambiguous annotations for all datasets. Futhermore, we build an evaluation system and carefully choose evaluations metrics\nfor every task, thus establishing fair comparison across Korean language models.\n", "citation": "@misc{park2021klue,\n      title={KLUE: Korean Language Understanding Evaluation},\n      author={Sungjoon Park and Jihyung Moon and Sungdong Kim and Won Ik Cho and Jiyoon Han and Jangwon Park and Chisung Song and Junseong Kim and Yongsook Song and Taehwan Oh and Joohong Lee and Juhyun Oh and Sungwon Lyu and Younghoon Jeong and Inkwon Lee and Sangwoo Seo and Dongjun Lee and Hyunwoo Kim and Myeonghwa Lee and Seongbo Jang and Seungwon Do and Sunkyoung Kim and Kyungtae Lim and Jongwon Lee and Kyumin Park and Jamin Shin and Seonghyun Kim and Lucy Park and Alice Oh and Jungwoo Ha and Kyunghyun Cho},\n      year={2021},\n      eprint={2105.09680},\n      archivePrefix={arXiv},\n      primaryClass={cs.CL}\n}\n", "homepage": "https://klue-benchmark.com/tasks/66/overview/description", "license": "CC-BY-SA-4.0", "features": {"guid": {"dtype": "string", "id": null, "_type": "Value"}, "title": {"dtype": "string", "id": null, "_type": "Value"}, "label": {"num_classes": 7, "names": ["IT\uacfc\ud559", "\uacbd\uc81c", "\uc0ac\ud68c", "\uc0dd\ud65c\ubb38\ud654", "\uc138\uacc4", "\uc2a4\ud3ec\uce20", "\uc815\uce58"], "names_file": null, "id": null, "_type": "ClassLabel"}, "url": {"dtype": "string", "id": null, "_type": "Value"}, "date": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "klue", "config_name": "ynat", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 10109664, "num_examples": 45678, "dataset_name": "klue"}, "validation": {"name": "validation", "num_bytes": 2039197, "num_examples": 9107, "dataset_name": "klue"}}, "download_checksums": {"http://klue-benchmark.com.s3.amazonaws.com/app/Competitions/000066/data/ynat-v1.tar.gz": {"num_bytes": 4932555, "checksum": "820a4d1d6d1fd83e2a421f856965d3cfc5c93627935ce8c5b27468c6113fc482"}}, "download_size": 4932555, "post_processing_size": null, "dataset_size": 12148861, "size_in_bytes": 17081416}, "sts": {"description": "KLUE (Korean Language Understanding Evaluation)\nKorean Language Understanding Evaluation (KLUE) benchmark is a series of datasets to evaluate natural language\nunderstanding capability of Korean language models. KLUE consists of 8 diverse and representative tasks, which are accessible\nto anyone without any restrictions. With ethical considerations in mind, we deliberately design annotation guidelines to obtain\nunambiguous annotations for all datasets. Futhermore, we build an evaluation system and carefully choose evaluations metrics\nfor every task, thus establishing fair comparison across Korean language models.\n", "citation": "@misc{park2021klue,\n      title={KLUE: Korean Language Understanding Evaluation},\n      author={Sungjoon Park and Jihyung Moon and Sungdong Kim and Won Ik Cho and Jiyoon Han and Jangwon Park and Chisung Song and Junseong Kim and Yongsook Song and Taehwan Oh and Joohong Lee and Juhyun Oh and Sungwon Lyu and Younghoon Jeong and Inkwon Lee and Sangwoo Seo and Dongjun Lee and Hyunwoo Kim and Myeonghwa Lee and Seongbo Jang and Seungwon Do and Sunkyoung Kim and Kyungtae Lim and Jongwon Lee and Kyumin Park and Jamin Shin and Seonghyun Kim and Lucy Park and Alice Oh and Jungwoo Ha and Kyunghyun Cho},\n      year={2021},\n      eprint={2105.09680},\n      archivePrefix={arXiv},\n      primaryClass={cs.CL}\n}\n", "homepage": "https://klue-benchmark.com/tasks/67/overview/description", "license": "CC-BY-SA-4.0", "features": {"guid": {"dtype": "string", "id": null, "_type": "Value"}, "source": {"dtype": "string", "id": null, "_type": "Value"}, "sentence1": {"dtype": "string", "id": null, "_type": "Value"}, "sentence2": {"dtype": "string", "id": null, "_type": "Value"}, "labels": {"label": {"dtype": "float64", "id": null, "_type": "Value"}, "real-label": {"dtype": "float64", "id": null, "_type": "Value"}, "binary-label": {"num_classes": 2, "names": ["negative", "positive"], "names_file": null, "id": null, "_type": "ClassLabel"}}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "klue", "config_name": "sts", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 2832921, "num_examples": 11668, "dataset_name": "klue"}, "validation": {"name": "validation", "num_bytes": 122657, "num_examples": 519, "dataset_name": "klue"}}, "download_checksums": {"http://klue-benchmark.com.s3.amazonaws.com/app/Competitions/000067/data/klue-sts-v1.tar.gz": {"num_bytes": 1349875, "checksum": "539341ba78a3b351c686cf70a448ac7a5886ed95f0719d5e3d2378ba703213bd"}}, "download_size": 1349875, "post_processing_size": null, "dataset_size": 2955578, "size_in_bytes": 4305453}, "nli": {"description": "KLUE (Korean Language Understanding Evaluation)\nKorean Language Understanding Evaluation (KLUE) benchmark is a series of datasets to evaluate natural language\nunderstanding capability of Korean language models. KLUE consists of 8 diverse and representative tasks, which are accessible\nto anyone without any restrictions. With ethical considerations in mind, we deliberately design annotation guidelines to obtain\nunambiguous annotations for all datasets. Futhermore, we build an evaluation system and carefully choose evaluations metrics\nfor every task, thus establishing fair comparison across Korean language models.\n", "citation": "@misc{park2021klue,\n      title={KLUE: Korean Language Understanding Evaluation},\n      author={Sungjoon Park and Jihyung Moon and Sungdong Kim and Won Ik Cho and Jiyoon Han and Jangwon Park and Chisung Song and Junseong Kim and Yongsook Song and Taehwan Oh and Joohong Lee and Juhyun Oh and Sungwon Lyu and Younghoon Jeong and Inkwon Lee and Sangwoo Seo and Dongjun Lee and Hyunwoo Kim and Myeonghwa Lee and Seongbo Jang and Seungwon Do and Sunkyoung Kim and Kyungtae Lim and Jongwon Lee and Kyumin Park and Jamin Shin and Seonghyun Kim and Lucy Park and Alice Oh and Jungwoo Ha and Kyunghyun Cho},\n      year={2021},\n      eprint={2105.09680},\n      archivePrefix={arXiv},\n      primaryClass={cs.CL}\n}\n", "homepage": "https://klue-benchmark.com/tasks/68/overview/description", "license": "CC-BY-SA-4.0", "features": {"guid": {"dtype": "string", "id": null, "_type": "Value"}, "source": {"dtype": "string", "id": null, "_type": "Value"}, "premise": {"dtype": "string", "id": null, "_type": "Value"}, "hypothesis": {"dtype": "string", "id": null, "_type": "Value"}, "label": {"num_classes": 3, "names": ["entailment", "neutral", "contradiction"], "names_file": null, "id": null, "_type": "ClassLabel"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "klue", "config_name": "nli", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 5719930, "num_examples": 24998, "dataset_name": "klue"}, "validation": {"name": "validation", "num_bytes": 673276, "num_examples": 3000, "dataset_name": "klue"}}, "download_checksums": {"http://klue-benchmark.com.s3.amazonaws.com/app/Competitions/000068/data/klue-nli-v1.tar.gz": {"num_bytes": 1257374, "checksum": "388be2033ef712072201903795a35b4f86826ee3ed3b62dc0c98e1721baa8850"}}, "download_size": 1257374, "post_processing_size": null, "dataset_size": 6393206, "size_in_bytes": 7650580}, "ner": {"description": "KLUE (Korean Language Understanding Evaluation)\nKorean Language Understanding Evaluation (KLUE) benchmark is a series of datasets to evaluate natural language\nunderstanding capability of Korean language models. KLUE consists of 8 diverse and representative tasks, which are accessible\nto anyone without any restrictions. With ethical considerations in mind, we deliberately design annotation guidelines to obtain\nunambiguous annotations for all datasets. Futhermore, we build an evaluation system and carefully choose evaluations metrics\nfor every task, thus establishing fair comparison across Korean language models.\n", "citation": "@misc{park2021klue,\n      title={KLUE: Korean Language Understanding Evaluation},\n      author={Sungjoon Park and Jihyung Moon and Sungdong Kim and Won Ik Cho and Jiyoon Han and Jangwon Park and Chisung Song and Junseong Kim and Yongsook Song and Taehwan Oh and Joohong Lee and Juhyun Oh and Sungwon Lyu and Younghoon Jeong and Inkwon Lee and Sangwoo Seo and Dongjun Lee and Hyunwoo Kim and Myeonghwa Lee and Seongbo Jang and Seungwon Do and Sunkyoung Kim and Kyungtae Lim and Jongwon Lee and Kyumin Park and Jamin Shin and Seonghyun Kim and Lucy Park and Alice Oh and Jungwoo Ha and Kyunghyun Cho},\n      year={2021},\n      eprint={2105.09680},\n      archivePrefix={arXiv},\n      primaryClass={cs.CL}\n}\n", "homepage": "https://klue-benchmark.com/tasks/69/overview/description", "license": "CC-BY-SA-4.0", "features": {"sentence": {"dtype": "string", "id": null, "_type": "Value"}, "tokens": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "ner_tags": {"feature": {"num_classes": 13, "names": ["B-DT", "I-DT", "B-LC", "I-LC", "B-OG", "I-OG", "B-PS", "I-PS", "B-QT", "I-QT", "B-TI", "I-TI", "O"], "names_file": null, "id": null, "_type": "ClassLabel"}, "length": -1, "id": null, "_type": "Sequence"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "klue", "config_name": "ner", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 19891953, "num_examples": 21008, "dataset_name": "klue"}, "validation": {"name": "validation", "num_bytes": 4937579, "num_examples": 5000, "dataset_name": "klue"}}, "download_checksums": {"http://klue-benchmark.com.s3.amazonaws.com/app/Competitions/000069/data/klue-ner-v1.tar.gz": {"num_bytes": 4308644, "checksum": "848a89759ac6b7c149c9a00d820726fe2a140c22782201f1a40d856672e7ea8e"}}, "download_size": 4308644, "post_processing_size": null, "dataset_size": 24829532, "size_in_bytes": 29138176}, "re": {"description": "KLUE (Korean Language Understanding Evaluation)\nKorean Language Understanding Evaluation (KLUE) benchmark is a series of datasets to evaluate natural language\nunderstanding capability of Korean language models. KLUE consists of 8 diverse and representative tasks, which are accessible\nto anyone without any restrictions. With ethical considerations in mind, we deliberately design annotation guidelines to obtain\nunambiguous annotations for all datasets. Futhermore, we build an evaluation system and carefully choose evaluations metrics\nfor every task, thus establishing fair comparison across Korean language models.\n", "citation": "@misc{park2021klue,\n      title={KLUE: Korean Language Understanding Evaluation},\n      author={Sungjoon Park and Jihyung Moon and Sungdong Kim and Won Ik Cho and Jiyoon Han and Jangwon Park and Chisung Song and Junseong Kim and Yongsook Song and Taehwan Oh and Joohong Lee and Juhyun Oh and Sungwon Lyu and Younghoon Jeong and Inkwon Lee and Sangwoo Seo and Dongjun Lee and Hyunwoo Kim and Myeonghwa Lee and Seongbo Jang and Seungwon Do and Sunkyoung Kim and Kyungtae Lim and Jongwon Lee and Kyumin Park and Jamin Shin and Seonghyun Kim and Lucy Park and Alice Oh and Jungwoo Ha and Kyunghyun Cho},\n      year={2021},\n      eprint={2105.09680},\n      archivePrefix={arXiv},\n      primaryClass={cs.CL}\n}\n", "homepage": "https://klue-benchmark.com/tasks/70/overview/description", "license": "CC-BY-SA-4.0", "features": {"guid": {"dtype": "string", "id": null, "_type": "Value"}, "sentence": {"dtype": "string", "id": null, "_type": "Value"}, "subject_entity": {"word": {"dtype": "string", "id": null, "_type": "Value"}, "start_idx": {"dtype": "int32", "id": null, "_type": "Value"}, "end_idx": {"dtype": "int32", "id": null, "_type": "Value"}, "type": {"dtype": "string", "id": null, "_type": "Value"}}, "object_entity": {"word": {"dtype": "string", "id": null, "_type": "Value"}, "start_idx": {"dtype": "int32", "id": null, "_type": "Value"}, "end_idx": {"dtype": "int32", "id": null, "_type": "Value"}, "type": {"dtype": "string", "id": null, "_type": "Value"}}, "label": {"num_classes": 30, "names": ["no_relation", "org:dissolved", "org:founded", "org:place_of_headquarters", "org:alternate_names", "org:member_of", "org:members", "org:political/religious_affiliation", "org:product", "org:founded_by", "org:top_members/employees", "org:number_of_employees/members", "per:date_of_birth", "per:date_of_death", "per:place_of_birth", "per:place_of_death", "per:place_of_residence", "per:origin", "per:employee_of", "per:schools_attended", "per:alternate_names", "per:parents", "per:children", "per:siblings", "per:spouse", "per:other_family", "per:colleagues", "per:product", "per:religion", "per:title"], "names_file": null, "id": null, "_type": "ClassLabel"}, "source": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "klue", "config_name": "re", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 11145538, "num_examples": 32470, "dataset_name": "klue"}, "validation": {"name": "validation", "num_bytes": 2559300, "num_examples": 7765, "dataset_name": "klue"}}, "download_checksums": {"http://klue-benchmark.com.s3.amazonaws.com/app/Competitions/000070/data/klue-re-v1.tar.gz": {"num_bytes": 5669259, "checksum": "b09ceac0d986cc09e42fcda9c7f2873c0eea8ec0629baf91fead36580790f8f5"}}, "download_size": 5669259, "post_processing_size": null, "dataset_size": 13704838, "size_in_bytes": 19374097}, "dp": {"description": "KLUE (Korean Language Understanding Evaluation)\nKorean Language Understanding Evaluation (KLUE) benchmark is a series of datasets to evaluate natural language\nunderstanding capability of Korean language models. KLUE consists of 8 diverse and representative tasks, which are accessible\nto anyone without any restrictions. With ethical considerations in mind, we deliberately design annotation guidelines to obtain\nunambiguous annotations for all datasets. Futhermore, we build an evaluation system and carefully choose evaluations metrics\nfor every task, thus establishing fair comparison across Korean language models.\n", "citation": "@misc{park2021klue,\n      title={KLUE: Korean Language Understanding Evaluation},\n      author={Sungjoon Park and Jihyung Moon and Sungdong Kim and Won Ik Cho and Jiyoon Han and Jangwon Park and Chisung Song and Junseong Kim and Yongsook Song and Taehwan Oh and Joohong Lee and Juhyun Oh and Sungwon Lyu and Younghoon Jeong and Inkwon Lee and Sangwoo Seo and Dongjun Lee and Hyunwoo Kim and Myeonghwa Lee and Seongbo Jang and Seungwon Do and Sunkyoung Kim and Kyungtae Lim and Jongwon Lee and Kyumin Park and Jamin Shin and Seonghyun Kim and Lucy Park and Alice Oh and Jungwoo Ha and Kyunghyun Cho},\n      year={2021},\n      eprint={2105.09680},\n      archivePrefix={arXiv},\n      primaryClass={cs.CL}\n}\n", "homepage": "https://klue-benchmark.com/tasks/71/overview/description", "license": "CC-BY-SA-4.0", "features": {"sentence": {"dtype": "string", "id": null, "_type": "Value"}, "index": [{"dtype": "int32", "id": null, "_type": "Value"}], "word_form": [{"dtype": "string", "id": null, "_type": "Value"}], "lemma": [{"dtype": "string", "id": null, "_type": "Value"}], "pos": [{"dtype": "string", "id": null, "_type": "Value"}], "head": [{"dtype": "int32", "id": null, "_type": "Value"}], "deprel": [{"dtype": "string", "id": null, "_type": "Value"}]}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "klue", "config_name": "dp", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 7900009, "num_examples": 10000, "dataset_name": "klue"}, "validation": {"name": "validation", "num_bytes": 1557506, "num_examples": 2000, "dataset_name": "klue"}}, "download_checksums": {"http://klue-benchmark.com.s3.amazonaws.com/app/Competitions/000071/data/klue-dp-v1.tar.gz": {"num_bytes": 2033461, "checksum": "2c76a3543a50599ac6640ad360ba00eac36e0b5b2363f708a614d6e50844d17b"}}, "download_size": 2033461, "post_processing_size": null, "dataset_size": 9457515, "size_in_bytes": 11490976}, "mrc": {"description": "KLUE (Korean Language Understanding Evaluation)\nKorean Language Understanding Evaluation (KLUE) benchmark is a series of datasets to evaluate natural language\nunderstanding capability of Korean language models. KLUE consists of 8 diverse and representative tasks, which are accessible\nto anyone without any restrictions. With ethical considerations in mind, we deliberately design annotation guidelines to obtain\nunambiguous annotations for all datasets. Futhermore, we build an evaluation system and carefully choose evaluations metrics\nfor every task, thus establishing fair comparison across Korean language models.\n", "citation": "@misc{park2021klue,\n      title={KLUE: Korean Language Understanding Evaluation},\n      author={Sungjoon Park and Jihyung Moon and Sungdong Kim and Won Ik Cho and Jiyoon Han and Jangwon Park and Chisung Song and Junseong Kim and Yongsook Song and Taehwan Oh and Joohong Lee and Juhyun Oh and Sungwon Lyu and Younghoon Jeong and Inkwon Lee and Sangwoo Seo and Dongjun Lee and Hyunwoo Kim and Myeonghwa Lee and Seongbo Jang and Seungwon Do and Sunkyoung Kim and Kyungtae Lim and Jongwon Lee and Kyumin Park and Jamin Shin and Seonghyun Kim and Lucy Park and Alice Oh and Jungwoo Ha and Kyunghyun Cho},\n      year={2021},\n      eprint={2105.09680},\n      archivePrefix={arXiv},\n      primaryClass={cs.CL}\n}\n", "homepage": "https://klue-benchmark.com/tasks/72/overview/description", "license": "CC-BY-SA-4.0", "features": {"title": {"dtype": "string", "id": null, "_type": "Value"}, "context": {"dtype": "string", "id": null, "_type": "Value"}, "news_category": {"dtype": "string", "id": null, "_type": "Value"}, "source": {"dtype": "string", "id": null, "_type": "Value"}, "guid": {"dtype": "string", "id": null, "_type": "Value"}, "is_impossible": {"dtype": "bool", "id": null, "_type": "Value"}, "question_type": {"dtype": "int32", "id": null, "_type": "Value"}, "question": {"dtype": "string", "id": null, "_type": "Value"}, "answers": {"feature": {"answer_start": {"dtype": "int32", "id": null, "_type": "Value"}, "text": {"dtype": "string", "id": null, "_type": "Value"}}, "length": -1, "id": null, "_type": "Sequence"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "klue", "config_name": "mrc", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 46505665, "num_examples": 17554, "dataset_name": "klue"}, "validation": {"name": "validation", "num_bytes": 15583053, "num_examples": 5841, "dataset_name": "klue"}}, "download_checksums": {"http://klue-benchmark.com.s3.amazonaws.com/app/Competitions/000072/data/klue-mrc-v1.tar.gz": {"num_bytes": 19218422, "checksum": "a444af252901452380d58a6320908ce4a86759bb6f38ad95d0ca98584ad33d14"}}, "download_size": 19218422, "post_processing_size": null, "dataset_size": 62088718, "size_in_bytes": 81307140}, "wos": {"description": "KLUE (Korean Language Understanding Evaluation)\nKorean Language Understanding Evaluation (KLUE) benchmark is a series of datasets to evaluate natural language\nunderstanding capability of Korean language models. KLUE consists of 8 diverse and representative tasks, which are accessible\nto anyone without any restrictions. With ethical considerations in mind, we deliberately design annotation guidelines to obtain\nunambiguous annotations for all datasets. Futhermore, we build an evaluation system and carefully choose evaluations metrics\nfor every task, thus establishing fair comparison across Korean language models.\n", "citation": "@misc{park2021klue,\n      title={KLUE: Korean Language Understanding Evaluation},\n      author={Sungjoon Park and Jihyung Moon and Sungdong Kim and Won Ik Cho and Jiyoon Han and Jangwon Park and Chisung Song and Junseong Kim and Yongsook Song and Taehwan Oh and Joohong Lee and Juhyun Oh and Sungwon Lyu and Younghoon Jeong and Inkwon Lee and Sangwoo Seo and Dongjun Lee and Hyunwoo Kim and Myeonghwa Lee and Seongbo Jang and Seungwon Do and Sunkyoung Kim and Kyungtae Lim and Jongwon Lee and Kyumin Park and Jamin Shin and Seonghyun Kim and Lucy Park and Alice Oh and Jungwoo Ha and Kyunghyun Cho},\n      year={2021},\n      eprint={2105.09680},\n      archivePrefix={arXiv},\n      primaryClass={cs.CL}\n}\n", "homepage": "https://klue-benchmark.com/tasks/73/overview/description", "license": "CC-BY-SA-4.0", "features": {"guid": {"dtype": "string", "id": null, "_type": "Value"}, "domains": [{"dtype": "string", "id": null, "_type": "Value"}], "dialogue": [{"role": {"dtype": "string", "id": null, "_type": "Value"}, "text": {"dtype": "string", "id": null, "_type": "Value"}, "state": [{"dtype": "string", "id": null, "_type": "Value"}]}]}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "klue", "config_name": "wos", "version": {"version_str": "1.0.0", "description": "", "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 26677002, "num_examples": 8000, "dataset_name": "klue"}, "validation": {"name": "validation", "num_bytes": 3488943, "num_examples": 1000, "dataset_name": "klue"}}, "download_checksums": {"http://klue-benchmark.com.s3.amazonaws.com/app/Competitions/000073/data/wos-v1.tar.gz": {"num_bytes": 4785657, "checksum": "da17829300271560afc6e7fc330503c2ca6f7ae7721d9bb94308579542a5871f"}}, "download_size": 4785657, "post_processing_size": null, "dataset_size": 30165945, "size_in_bytes": 34951602}}

dummy/dp/1.0.0/dummy_data.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f68428f7376efdd7b1ced3a41e2276ea5b78af993c33ff6437ea37478fe150d7
+size 2085

dummy/mrc/1.0.0/dummy_data.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:608f06472da3a89a64a9e60a926a6d61c5b6bf5e8b7da5c12f74168bf45afe70
+size 3306

dummy/ner/1.0.0/dummy_data.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b13d84876433d36a7cd6a063e1140427046ec46e81a2a771135860789493b32a
+size 1950

dummy/nli/1.0.0/dummy_data.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0a19d1a92e3acc1e68e435243199e2fca98d1e5a8fdb397315b924f7f122c31b
+size 1767

dummy/re/1.0.0/dummy_data.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ea6a2fabc3378ba3dc8e6b233ae2c11d83642d72ff44a345ed7a66cdddfccda9
+size 2982

dummy/sts/1.0.0/dummy_data.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a176a30bde7e52eb5c6d47f5f3a210ca1150b1c59c122c38d487b60f9c8daa50
+size 2790

dummy/wos/1.0.0/dummy_data.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c1e736318e979928cdbe1d69a751f436aac60b9d503707b369ae2d1352162c1a
+size 6578

dummy/ynat/1.0.0/dummy_data.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d2f02f6e8ebe7660d3548d53f27b9382d0301a39e3c08c1872db5513ee996c2a
+size 2400

klue.py

@@ -0,0 +1,521 @@
+# coding=utf-8
+# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""KLUE (Korean Language Understanding Evaluation) benchmark."""
+
+
+import csv
+import json
+import os
+import textwrap
+
+import datasets
+
+
+_KLUE_CITATION = """\
+@misc{park2021klue,
+      title={KLUE: Korean Language Understanding Evaluation},
+      author={Sungjoon Park and Jihyung Moon and Sungdong Kim and Won Ik Cho and Jiyoon Han and Jangwon Park and Chisung Song and Junseong Kim and Yongsook Song and Taehwan Oh and Joohong Lee and Juhyun Oh and Sungwon Lyu and Younghoon Jeong and Inkwon Lee and Sangwoo Seo and Dongjun Lee and Hyunwoo Kim and Myeonghwa Lee and Seongbo Jang and Seungwon Do and Sunkyoung Kim and Kyungtae Lim and Jongwon Lee and Kyumin Park and Jamin Shin and Seonghyun Kim and Lucy Park and Alice Oh and Jungwoo Ha and Kyunghyun Cho},
+      year={2021},
+      eprint={2105.09680},
+      archivePrefix={arXiv},
+      primaryClass={cs.CL}
+}
+"""
+
+_KLUE_DESCRIPTION = """\
+KLUE (Korean Language Understanding Evaluation)
+Korean Language Understanding Evaluation (KLUE) benchmark is a series of datasets to evaluate natural language
+understanding capability of Korean language models. KLUE consists of 8 diverse and representative tasks, which are accessible
+to anyone without any restrictions. With ethical considerations in mind, we deliberately design annotation guidelines to obtain
+unambiguous annotations for all datasets. Futhermore, we build an evaluation system and carefully choose evaluations metrics
+for every task, thus establishing fair comparison across Korean language models.
+"""
+
+_DATA_URLs = {
+    "ynat": "http://klue-benchmark.com.s3.amazonaws.com/app/Competitions/000066/data/ynat-v1.tar.gz",
+    "sts": "http://klue-benchmark.com.s3.amazonaws.com/app/Competitions/000067/data/klue-sts-v1.tar.gz",
+    "nli": "http://klue-benchmark.com.s3.amazonaws.com/app/Competitions/000068/data/klue-nli-v1.tar.gz",
+    "ner": "http://klue-benchmark.com.s3.amazonaws.com/app/Competitions/000069/data/klue-ner-v1.tar.gz",
+    "re": "http://klue-benchmark.com.s3.amazonaws.com/app/Competitions/000070/data/klue-re-v1.tar.gz",
+    "dp": "http://klue-benchmark.com.s3.amazonaws.com/app/Competitions/000071/data/klue-dp-v1.tar.gz",
+    "mrc": "http://klue-benchmark.com.s3.amazonaws.com/app/Competitions/000072/data/klue-mrc-v1.tar.gz",
+    "wos": "http://klue-benchmark.com.s3.amazonaws.com/app/Competitions/000073/data/wos-v1.tar.gz",
+}
+
+_DESCRIPTION_URLs = {
+    "ynat": "https://klue-benchmark.com/tasks/66/overview/description",
+    "sts": "https://klue-benchmark.com/tasks/67/overview/description",
+    "nli": "https://klue-benchmark.com/tasks/68/overview/description",
+    "ner": "https://klue-benchmark.com/tasks/69/overview/description",
+    "re": "https://klue-benchmark.com/tasks/70/overview/description",
+    "dp": "https://klue-benchmark.com/tasks/71/overview/description",
+    "mrc": "https://klue-benchmark.com/tasks/72/overview/description",
+    "wos": "https://klue-benchmark.com/tasks/73/overview/description",
+}
+
+_LICENSE = "CC-BY-SA-4.0"
+
+
+class KlueConfig(datasets.BuilderConfig):
+    """BuilderConfig for KLUE."""
+
+    def __init__(
+        self,
+        features,
+        data_url,
+        url,
+        file_map,
+        **kwargs,
+    ):
+        """BuilderConfig for KLUE."""
+
+        super(KlueConfig, self).__init__(version=datasets.Version("1.0.0", ""), **kwargs)
+        self.features = features
+        self.data_url = data_url
+        self.url = url
+        self.file_map = file_map
+
+
+class Klue(datasets.GeneratorBasedBuilder):
+    """The General Language Understanding Evaluation (GLUE) benchmark."""
+
+    BUILDER_CONFIGS = [
+        KlueConfig(
+            name="ynat",
+            features={
+                "guid": datasets.Value("string"),
+                "title": datasets.Value("string"),
+                "label": datasets.features.ClassLabel(names=["IT과학", "경제", "사회", "생활문화", "세계", "스포츠", "정치"]),
+                "url": datasets.Value("string"),
+                "date": datasets.Value("string"),
+            },
+            description=textwrap.dedent(
+                """\
+            In topic classification (TC), the goal is to predict the topic of a given text
+            snippet. We include TC in our KLUE benchmark, as inferring the topic of a text is a key
+            capability that should be possessed by a language understanding system.
+            Following a typical single  sentence classification task, we introduce YNAT, a Younhap
+            News Agency news headlines for Topic Classification. For Korean, no dataset has been
+            proposed for this task, which motivates us to construct the first Korean topic
+            classification benchmark. In this task, given a news headline, a text classifier must
+            predict a topic which is one of politics, economy, society, culture, world, IT/science,
+            and sports. Macro-F1 score is used to evaluate a system."""
+            ),
+            data_url=_DATA_URLs["ynat"],
+            url=_DESCRIPTION_URLs["ynat"],
+            file_map={
+                "train": "ynat-v1_train.json",
+                "dev": "ynat-v1_dev.json",
+            },
+        ),
+        KlueConfig(
+            name="sts",
+            features={
+                "guid": datasets.Value("string"),
+                "source": datasets.Value("string"),
+                "sentence1": datasets.Value("string"),
+                "sentence2": datasets.Value("string"),
+                "labels": {
+                    "label": datasets.Value("float64"),
+                    "real-label": datasets.Value("float64"),
+                    "binary-label": datasets.ClassLabel(names=["negative", "positive"]),
+                },
+            },
+            description=textwrap.dedent(
+                """\
+            STS is a task which aims to predict the semantic similarity of two input sentences as
+            a real value between 0 and 5. Note that we furthure binarized the prediction scores
+            into two classes with a threshold score 3.0 (paraphrased or not) and evaluated with
+            a classification metric.
+            """
+            ),
+            data_url=_DATA_URLs["sts"],
+            url=_DESCRIPTION_URLs["sts"],
+            file_map={
+                "train": "klue-sts-v1_train.json",
+                "dev": "klue-sts-v1_dev.json",
+            },
+        ),
+        KlueConfig(
+            name="nli",
+            features={
+                "guid": datasets.Value("string"),
+                "source": datasets.Value("string"),
+                "premise": datasets.Value("string"),
+                "hypothesis": datasets.Value("string"),
+                "label": datasets.ClassLabel(names=["entailment", "neutral", "contradiction"]),
+            },
+            description=textwrap.dedent(
+                """\
+            NLI is a task to infer the relationship between a hypothesis sentence and a premise
+            sentence. Given the premise, the model determines if the hypothesis is true (entailment),
+            false (contradiction), or undetermined (neutral).
+            """
+            ),
+            data_url=_DATA_URLs["nli"],
+            url=_DESCRIPTION_URLs["nli"],
+            file_map={
+                "train": "klue-nli-v1_train.json",
+                "dev": "klue-nli-v1_dev.json",
+            },
+        ),
+        KlueConfig(
+            name="ner",
+            features={
+                "sentence": datasets.Value("string"),
+                "tokens": datasets.Sequence(datasets.Value("string")),
+                "ner_tags": datasets.Sequence(
+                    datasets.ClassLabel(
+                        names=[
+                            "B-DT",
+                            "I-DT",
+                            "B-LC",
+                            "I-LC",
+                            "B-OG",
+                            "I-OG",
+                            "B-PS",
+                            "I-PS",
+                            "B-QT",
+                            "I-QT",
+                            "B-TI",
+                            "I-TI",
+                            "O",
+                        ]
+                    )
+                ),
+            },
+            description=textwrap.dedent(
+                """\
+            NER is a task to detect the boundaries of named entities in unstructured text and to
+            classify the types. A named entity can be of one of predefined entity types such as
+            person, location, organization, time expressions, quantities and monetary values.
+            """
+            ),
+            data_url=_DATA_URLs["ner"],
+            url=_DESCRIPTION_URLs["ner"],
+            file_map={
+                "train": "klue-ner-v1_train.tsv",
+                "dev": "klue-ner-v1_dev.tsv",
+            },
+        ),
+        KlueConfig(
+            name="re",
+            features={
+                "guid": datasets.Value("string"),
+                "sentence": datasets.Value("string"),
+                "subject_entity": {
+                    "word": datasets.Value("string"),
+                    "start_idx": datasets.Value("int32"),
+                    "end_idx": datasets.Value("int32"),
+                    "type": datasets.Value("string"),
+                },
+                "object_entity": {
+                    "word": datasets.Value("string"),
+                    "start_idx": datasets.Value("int32"),
+                    "end_idx": datasets.Value("int32"),
+                    "type": datasets.Value("string"),
+                },
+                "label": datasets.ClassLabel(
+                    names=[
+                        "no_relation",
+                        "org:dissolved",
+                        "org:founded",
+                        "org:place_of_headquarters",
+                        "org:alternate_names",
+                        "org:member_of",
+                        "org:members",
+                        "org:political/religious_affiliation",
+                        "org:product",
+                        "org:founded_by",
+                        "org:top_members/employees",
+                        "org:number_of_employees/members",
+                        "per:date_of_birth",
+                        "per:date_of_death",
+                        "per:place_of_birth",
+                        "per:place_of_death",
+                        "per:place_of_residence",
+                        "per:origin",
+                        "per:employee_of",
+                        "per:schools_attended",
+                        "per:alternate_names",
+                        "per:parents",
+                        "per:children",
+                        "per:siblings",
+                        "per:spouse",
+                        "per:other_family",
+                        "per:colleagues",
+                        "per:product",
+                        "per:religion",
+                        "per:title",
+                    ]
+                ),
+                "source": datasets.Value("string"),
+            },
+            description=textwrap.dedent(
+                """\
+            RE is a task to identify semantic relations between entity pairs in a text. The relation
+            is defined between an entity pair consisting of subject entity and object entity.
+            The goal is then to pick an appropriate relationship between these two entities.
+            """
+            ),
+            data_url=_DATA_URLs["re"],
+            url=_DESCRIPTION_URLs["re"],
+            file_map={
+                "train": "klue-re-v1_train.json",
+                "dev": "klue-re-v1_dev.json",
+            },
+        ),
+        KlueConfig(
+            name="dp",
+            features={
+                "sentence": datasets.Value("string"),
+                "index": [datasets.Value("int32")],
+                "word_form": [datasets.Value("string")],
+                "lemma": [datasets.Value("string")],
+                "pos": [datasets.Value("string")],
+                "head": [datasets.Value("int32")],
+                "deprel": [datasets.Value("string")],
+            },
+            description=textwrap.dedent(
+                """\
+            DP is a task that aims at finding relational information among words.
+            The goal is to predict a graph structure and a dependency label of an input sentence
+            based on the dependency grammar.
+            """
+            ),
+            data_url=_DATA_URLs["dp"],
+            url=_DESCRIPTION_URLs["dp"],
+            file_map={
+                "train": "klue-dp-v1_train.tsv",
+                "dev": "klue-dp-v1_dev.tsv",
+            },
+        ),
+        KlueConfig(
+            name="mrc",
+            features={
+                "title": datasets.Value("string"),
+                "context": datasets.Value("string"),
+                "news_category": datasets.Value("string"),
+                "source": datasets.Value("string"),
+                "guid": datasets.Value("string"),
+                "is_impossible": datasets.Value("bool"),
+                "question_type": datasets.Value("int32"),
+                "question": datasets.Value("string"),
+                "answers": datasets.features.Sequence(
+                    {
+                        "answer_start": datasets.Value("int32"),
+                        "text": datasets.Value("string"),
+                    },
+                ),
+            },
+            description=textwrap.dedent(
+                """\
+            MRC is a task of evaluating model that can answer a question about a given text
+            passage. Specifically, we formulate the task as a span prediction task, where the
+            answer is a text segment (coined as spans) in the passage.
+            """
+            ),
+            data_url=_DATA_URLs["mrc"],
+            url=_DESCRIPTION_URLs["mrc"],
+            file_map={
+                "train": "klue-mrc-v1_train.json",
+                "dev": "klue-mrc-v1_dev.json",
+            },
+        ),
+        KlueConfig(
+            name="wos",
+            features={
+                "guid": datasets.Value("string"),
+                "domains": [datasets.Value("string")],
+                "dialogue": [
+                    {
+                        "role": datasets.Value("string"),
+                        "text": datasets.Value("string"),
+                        "state": [datasets.Value("string")],
+                    }
+                ],
+            },
+            description=textwrap.dedent(
+                """\
+            DST is a task to predict slot and value pairs (dialogue states) from a task-oriented
+            dialogue. The potential pairs are predefined by a given task schema and knowledge
+            base (KB).
+            """
+            ),
+            data_url=_DATA_URLs["wos"],
+            url=_DESCRIPTION_URLs["wos"],
+            file_map={
+                "train": "wos-v1_train.json",
+                "dev": "wos-v1_dev.json",
+            },
+        ),
+    ]
+
+    def _info(self):
+        return datasets.DatasetInfo(
+            description=_KLUE_DESCRIPTION,
+            features=datasets.Features(self.config.features),
+            homepage=self.config.url,
+            citation=_KLUE_CITATION,
+            license=_LICENSE,
+        )
+
+    def _split_generators(self, dl_manager):
+        dl_dir = dl_manager.download_and_extract(self.config.data_url)
+        dir_name = self.config.data_url.split("/")[-1].replace(".tar.gz", "")
+        data_dir = os.path.join(dl_dir, dir_name)
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN,
+                gen_kwargs={
+                    "data_file": os.path.join(data_dir, self.config.file_map["train"]),
+                    "split": "train",
+                },
+            ),
+            datasets.SplitGenerator(
+                name=datasets.Split.VALIDATION,
+                gen_kwargs={
+                    "data_file": os.path.join(data_dir, self.config.file_map["dev"]),
+                    "split": "dev",
+                },
+            ),
+        ]
+
+    def _generate_examples(self, data_file, split):
+        if self.config.name in ["ynat", "sts", "re"]:
+            with open(data_file, encoding="UTF-8") as f:
+                f = json.load(f)
+                for id_, row in enumerate(f):
+                    features = {key: row[key] for key in row if key in self.config.features}
+                    yield id_, features
+
+        if self.config.name == "nli":
+            with open(data_file, encoding="UTF-8") as f:
+                f = json.load(f)
+                for id_, row in enumerate(f):
+                    # In train file, "source" is written as "genre"
+                    features = {
+                        "guid": row["guid"],
+                        "source": row["source"] if "source" in row else row["genre"],
+                        "premise": row["premise"],
+                        "hypothesis": row["hypothesis"],
+                        "label": row["gold_label"],
+                    }
+                    yield id_, features
+
+        if self.config.name == "ner":
+            with open(data_file, encoding="UTF-8") as f:
+                reader = csv.reader(f, delimiter="\t", quoting=csv.QUOTE_NONE)
+                for _ in range(5):  # skip headers
+                    next(reader)
+                id_ = -1
+                for row in reader:
+                    if row:
+                        if row[0].startswith("##"):
+                            id_ += 1
+                            tokens, ner_tags = [], []
+                            sentence = row[1]
+                        else:
+                            tokens.append(row[0])
+                            ner_tags.append(row[1])
+                    else:  # new line
+                        assert len(tokens) == len(ner_tags)
+                        yield id_, {"sentence": sentence, "tokens": tokens, "ner_tags": ner_tags}
+
+        if self.config.name == "dp":
+            with open(data_file, encoding="UTF-8") as f:
+                reader = csv.reader(f, delimiter="\t", quoting=csv.QUOTE_NONE)
+                for _ in range(5):  # skip headers
+                    next(reader)
+                id_ = -1
+                for row in reader:
+                    if row:
+                        if row[0].startswith("##"):
+                            id_ += 1
+                            index = []
+                            word_form = []
+                            lemma = []
+                            pos = []
+                            head = []
+                            deprel = []
+                            sentence = row[1]
+                        else:
+                            index.append(row[0])
+                            word_form.append(row[1])
+                            lemma.append(row[2])
+                            pos.append(row[3])
+                            head.append(row[4])
+                            deprel.append(row[5])
+                    else:  # new line
+                        assert len(index) == len(word_form) == len(lemma) == len(pos) == len(head) == len(deprel)
+                        yield id_, {
+                            "sentence": sentence,
+                            "index": index,
+                            "word_form": word_form,
+                            "lemma": lemma,
+                            "pos": pos,
+                            "head": head,
+                            "deprel": deprel,
+                        }
+
+        if self.config.name == "mrc":
+            with open(data_file, encoding="UTF-8") as f:
+                f = json.load(f)
+                id_ = -1
+                for example in f["data"]:
+                    title = example.get("title", "")
+                    news_category = example.get("news_category", "")
+                    source = example["source"]
+                    for paragraph in example["paragraphs"]:
+                        context = paragraph["context"].strip()
+                        for qa in paragraph["qas"]:
+                            guid = qa["guid"]
+                            question_type = qa["question_type"]
+                            is_impossible = qa["is_impossible"]
+                            question = qa["question"].strip()
+
+                            if "plausible_answers" in qa:
+                                qa["answers"].extend(qa["plausible_answers"])
+                            answer_starts = [answer["answer_start"] for answer in qa["answers"]]
+                            answers = [answer["text"].strip() for answer in qa["answers"]]
+                            id_ += 1
+
+                            yield id_, {
+                                "guid": guid,
+                                "title": title,
+                                "context": context,
+                                "news_category": news_category,
+                                "source": source,
+                                "question_type": question_type,
+                                "is_impossible": is_impossible,
+                                "question": question,
+                                "answers": {
+                                    "answer_start": answer_starts,
+                                    "text": answers,
+                                },
+                            }
+
+        if self.config.name == "wos":
+            with open(data_file, encoding="UTF-8") as f:
+                f = json.load(f)
+                for id_, row in enumerate(f):
+                    guid = row["guid"]
+                    domains = row["domains"]
+                    dialogue = row["dialogue"]
+                    for utterance in dialogue:
+                        if "state" not in utterance:
+                            utterance["state"] = []
+                    yield id_, {"guid": guid, "domains": domains, "dialogue": dialogue}