эснли

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

The e-SNLI dataset extends the Stanford Natural Language Inference Dataset to
include human-annotated natural language explanations of the entailment
relations.

{
    "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"
    },
    "explanation_1": {
        "dtype": "string",
        "id": null,
        "_type": "Value"
    },
    "explanation_2": {
        "dtype": "string",
        "id": null,
        "_type": "Value"
    },
    "explanation_3": {
        "dtype": "string",
        "id": null,
        "_type": "Value"
    }
}