kuora

ds = tfds.load('huggingface:quora')

The Quora dataset is composed of question pairs, and the task is to determine if the questions are paraphrases of each other (have the same meaning).

{
    "questions": {
        "feature": {
            "id": {
                "dtype": "int32",
                "id": null,
                "_type": "Value"
            },
            "text": {
                "dtype": "string",
                "id": null,
                "_type": "Value"
            }
        },
        "length": -1,
        "id": null,
        "_type": "Sequence"
    },
    "is_duplicate": {
        "dtype": "bool",
        "id": null,
        "_type": "Value"
    }
}