Datasets:

Polygl0t
/

CALAME-PT

dataset_info:
  - config_name: pt
    features:
      - name: text
        dtype: string
    splits:
      - name: test
        num_bytes: 1273606
        num_examples: 2076
    download_size: 490275
    dataset_size: 1273606
configs:
  - config_name: pt
    default: true
    data_files:
      - split: test
        path: pt/test-*
license: mit
language:
  - pt
tags:
  - generation
  - language modeling
  - portuguese
  - lambada
  - zero-shot
size_categories:
  - 1K<n<10K

Context-Aware LAnguage Modeling Evaluation for Portuguese (CALAME-PT)

This is a version of the original CALAME dataset adapted for use using the LM Evaluation Harness.

CALAME-PT is a PT benchmark composed of small texts (contexts) and their respective last words.

These contexts should, in theory, contain enough information so that a human or a model is capable of guessing its last word - without being too specific and/or too ambiguous.

Citation

All credits to the original CALAME dataset creators. Please cite the following paper when using this dataset:

@article{lopes2024gl,
  title={Gl$\backslash$'orIA--A Generative and Open Large Language Model for Portuguese},
  author={Lopes, Ricardo and Magalh{\~a}es, Jo{\~a}o and Semedo, David},
  journal={arXiv preprint arXiv:2402.12969},
  year={2024}