References:
squad_v2
Use the following command to load this dataset in TFDS:
ds = tfds.load('huggingface:squad_v2/squad_v2')
- Description:
combines the 100,000 questions in SQuAD1.1 with over 50,000 unanswerable questions written adversarially by crowdworkers
to look similar to answerable ones. To do well on SQuAD2.0, systems must not only answer questions when possible, but
also determine when no answer is supported by the paragraph and abstain from answering.
- License: No known license
- Version: 2.0.0
- Splits:
| Split | Examples |
|---|---|
'train' |
130319 |
'validation' |
11873 |
- Features:
{
"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"
}
}