anli

ds = tfds.load('huggingface:anli/plain_text')

The Adversarial Natural Language Inference (ANLI) is a new large-scale NLI benchmark dataset, 
The dataset is collected via an iterative, adversarial human-and-model-in-the-loop procedure.
ANLI is much more difficult than its predecessors including SNLI and MNLI.
It contains three rounds. Each round has train/dev/test splits.

{
    "uid": {
        "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"
    },
    "reason": {
        "dtype": "string",
        "id": null,
        "_type": "Value"
    }
}