Attend the Women in ML Symposium on December 7 Register now

ai2_arc

References:

ARC-Challenge

Use the following command to load this dataset in TFDS:

ds = tfds.load('huggingface:ai2_arc/ARC-Challenge')
  • Description:
A new dataset of 7,787 genuine grade-school level, multiple-choice science questions, assembled to encourage research in
 advanced question-answering. The dataset is partitioned into a Challenge Set and an Easy Set, where the former contains
 only questions answered incorrectly by both a retrieval-based algorithm and a word co-occurrence algorithm. We are also
 including a corpus of over 14 million science sentences relevant to the task, and an implementation of three neural baseline models for this dataset. We pose ARC as a challenge to the community.
  • License: No known license
  • Version: 1.0.0
  • Splits:
Split Examples
'test' 1172
'train' 1119
'validation' 299
  • Features:
{
    "id": {
        "dtype": "string",
        "id": null,
        "_type": "Value"
    },
    "question": {
        "dtype": "string",
        "id": null,
        "_type": "Value"
    },
    "choices": {
        "feature": {
            "text": {
                "dtype": "string",
                "id": null,
                "_type": "Value"
            },
            "label": {
                "dtype": "string",
                "id": null,
                "_type": "Value"
            }
        },
        "length": -1,
        "id": null,
        "_type": "Sequence"
    },
    "answerKey": {
        "dtype": "string",
        "id": null,
        "_type": "Value"
    }
}

ARC-Easy

Use the following command to load this dataset in TFDS:

ds = tfds.load('huggingface:ai2_arc/ARC-Easy')
  • Description:
A new dataset of 7,787 genuine grade-school level, multiple-choice science questions, assembled to encourage research in
 advanced question-answering. The dataset is partitioned into a Challenge Set and an Easy Set, where the former contains
 only questions answered incorrectly by both a retrieval-based algorithm and a word co-occurrence algorithm. We are also
 including a corpus of over 14 million science sentences relevant to the task, and an implementation of three neural baseline models for this dataset. We pose ARC as a challenge to the community.
  • License: No known license
  • Version: 1.0.0
  • Splits:
Split Examples
'test' 2376
'train' 2251
'validation' 570
  • Features:
{
    "id": {
        "dtype": "string",
        "id": null,
        "_type": "Value"
    },
    "question": {
        "dtype": "string",
        "id": null,
        "_type": "Value"
    },
    "choices": {
        "feature": {
            "text": {
                "dtype": "string",
                "id": null,
                "_type": "Value"
            },
            "label": {
                "dtype": "string",
                "id": null,
                "_type": "Value"
            }
        },
        "length": -1,
        "id": null,
        "_type": "Sequence"
    },
    "answerKey": {
        "dtype": "string",
        "id": null,
        "_type": "Value"
    }
}