참고자료:
적대적QA
TFDS에 이 데이터세트를 로드하려면 다음 명령어를 사용하세요.
ds = tfds.load('huggingface:adversarial_qa/adversarialQA')
- 설명 :
AdversarialQA is a Reading Comprehension dataset, consisting of questions posed by crowdworkers on a set of Wikipedia articles using an adversarial model-in-the-loop.
We use three different models; BiDAF (Seo et al., 2016), BERT-Large (Devlin et al., 2018), and RoBERTa-Large (Liu et al., 2019) in the annotation loop and construct three datasets; D(BiDAF), D(BERT), and D(RoBERTa), each with 10,000 training examples, 1,000 validation, and 1,000 test examples.
The adversarial human annotation paradigm ensures that these datasets consist of questions that current state-of-the-art models (at least the ones used as adversaries in the annotation loop) find challenging.
- 라이센스 : 알려진 라이센스 없음
- 버전 : 1.0.0
- 분할 :
| 나뉘다 | 예 |
|---|---|
'test' | 3000 |
'train' | 30000 |
'validation' | 3000 |
- 특징 :
{
"id": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"title": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"context": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"question": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"answers": {
"feature": {
"text": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"answer_start": {
"dtype": "int32",
"id": null,
"_type": "Value"
}
},
"length": -1,
"id": null,
"_type": "Sequence"
},
"metadata": {
"split": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"model_in_the_loop": {
"dtype": "string",
"id": null,
"_type": "Value"
}
}
}
드비다프
TFDS에 이 데이터세트를 로드하려면 다음 명령어를 사용하세요.
ds = tfds.load('huggingface:adversarial_qa/dbidaf')
- 설명 :
AdversarialQA is a Reading Comprehension dataset, consisting of questions posed by crowdworkers on a set of Wikipedia articles using an adversarial model-in-the-loop.
We use three different models; BiDAF (Seo et al., 2016), BERT-Large (Devlin et al., 2018), and RoBERTa-Large (Liu et al., 2019) in the annotation loop and construct three datasets; D(BiDAF), D(BERT), and D(RoBERTa), each with 10,000 training examples, 1,000 validation, and 1,000 test examples.
The adversarial human annotation paradigm ensures that these datasets consist of questions that current state-of-the-art models (at least the ones used as adversaries in the annotation loop) find challenging.
- 라이센스 : 알려진 라이센스 없음
- 버전 : 1.0.0
- 분할 :
| 나뉘다 | 예 |
|---|---|
'test' | 1000 |
'train' | 10000 |
'validation' | 1000 |
- 특징 :
{
"id": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"title": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"context": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"question": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"answers": {
"feature": {
"text": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"answer_start": {
"dtype": "int32",
"id": null,
"_type": "Value"
}
},
"length": -1,
"id": null,
"_type": "Sequence"
},
"metadata": {
"split": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"model_in_the_loop": {
"dtype": "string",
"id": null,
"_type": "Value"
}
}
}
드버트
TFDS에 이 데이터세트를 로드하려면 다음 명령어를 사용하세요.
ds = tfds.load('huggingface:adversarial_qa/dbert')
- 설명 :
AdversarialQA is a Reading Comprehension dataset, consisting of questions posed by crowdworkers on a set of Wikipedia articles using an adversarial model-in-the-loop.
We use three different models; BiDAF (Seo et al., 2016), BERT-Large (Devlin et al., 2018), and RoBERTa-Large (Liu et al., 2019) in the annotation loop and construct three datasets; D(BiDAF), D(BERT), and D(RoBERTa), each with 10,000 training examples, 1,000 validation, and 1,000 test examples.
The adversarial human annotation paradigm ensures that these datasets consist of questions that current state-of-the-art models (at least the ones used as adversaries in the annotation loop) find challenging.
- 라이센스 : 알려진 라이센스 없음
- 버전 : 1.0.0
- 분할 :
| 나뉘다 | 예 |
|---|---|
'test' | 1000 |
'train' | 10000 |
'validation' | 1000 |
- 특징 :
{
"id": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"title": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"context": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"question": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"answers": {
"feature": {
"text": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"answer_start": {
"dtype": "int32",
"id": null,
"_type": "Value"
}
},
"length": -1,
"id": null,
"_type": "Sequence"
},
"metadata": {
"split": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"model_in_the_loop": {
"dtype": "string",
"id": null,
"_type": "Value"
}
}
}
드로베르타
TFDS에 이 데이터세트를 로드하려면 다음 명령어를 사용하세요.
ds = tfds.load('huggingface:adversarial_qa/droberta')
- 설명 :
AdversarialQA is a Reading Comprehension dataset, consisting of questions posed by crowdworkers on a set of Wikipedia articles using an adversarial model-in-the-loop.
We use three different models; BiDAF (Seo et al., 2016), BERT-Large (Devlin et al., 2018), and RoBERTa-Large (Liu et al., 2019) in the annotation loop and construct three datasets; D(BiDAF), D(BERT), and D(RoBERTa), each with 10,000 training examples, 1,000 validation, and 1,000 test examples.
The adversarial human annotation paradigm ensures that these datasets consist of questions that current state-of-the-art models (at least the ones used as adversaries in the annotation loop) find challenging.
- 라이센스 : 알려진 라이센스 없음
- 버전 : 1.0.0
- 분할 :
| 나뉘다 | 예 |
|---|---|
'test' | 1000 |
'train' | 10000 |
'validation' | 1000 |
- 특징 :
{
"id": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"title": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"context": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"question": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"answers": {
"feature": {
"text": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"answer_start": {
"dtype": "int32",
"id": null,
"_type": "Value"
}
},
"length": -1,
"id": null,
"_type": "Sequence"
},
"metadata": {
"split": {
"dtype": "string",
"id": null,
"_type": "Value"
},
"model_in_the_loop": {
"dtype": "string",
"id": null,
"_type": "Value"
}
}
}