---
dataset_info:
  features:
  - name: hypothesis
    dtype: string
  - name: gold_label
    dtype: string
  - name: explanation
    dtype: string
  - name: image
    dtype: image
  splits:
  - name: train
    num_bytes: 14399844155.0
    num_examples: 100000
  - name: validation
    num_bytes: 1997715632.232
    num_examples: 14339
  - name: test
    num_bytes: 2022573639.84
    num_examples: 14740
  download_size: 4053386669
  dataset_size: 18420133427.072
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
---

This is the official ESNLI-VE dataset with all the train, test, val splits. And I dont own the copyright of the work. All copyright of this dataset belongs to the original author of the work. Uploaded here so that community members can easily access this dataset and evaluate the models.

Original Paper link: https://arxiv.org/abs/1812.01193

We have also used this dataset in one of our work, published in EMNLP 2025, you can refer to it below:

Paper Link: https://arxiv.org/abs/2508.19724