README.md

metadata

dataset_info:
  features:
    - name: premise
      dtype: string
    - name: hypothesis
      dtype: string
    - name: label
      dtype: int64
    - name: id
      dtype: string
  splits:
    - name: train
      num_bytes: 74977588.8
      num_examples: 502724
    - name: test
      num_bytes: 18744397.2
      num_examples: 125681
  download_size: 58272954
  dataset_size: 93721986
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*

Norsk Bokmål NLI dataset

Machine translation of MNLI and SNLI to Bokmål. Based on tollefj/all-nli-NOB, but all neutral examples are removed, test-train split is done, and entailment is mapped to 1, while contradiction is mapped to 0. This is done so that we can use AnglE Training on the dataset.