---
license: apache-2.0
dataset_info:
  features:
  - name: inputs
    dtype: string
  - name: targets
    dtype: string
  - name: language
    dtype: string
  - name: language_code
    dtype: string
  - name: annotation_type
    dtype: string
  - name: user_id
    dtype: string
  splits:
  - name: train
    num_bytes: 11918662
    num_examples: 6633
  - name: test
    num_bytes: 308222
    num_examples: 250
  download_size: 4492541
  dataset_size: 12226884
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
task_categories:
- question-answering
language:
- bn
pretty_name: (Subset of) Aya dataset translated to Bengali
size_categories:
- 1K<n<10K
---


`aya_dataset_ben_translated` is a subset of the [aya_dataset](https://huggingface.co/datasets/CohereForAI/aya_dataset), with some modifications. In particular, the original data points in Bengali (indicated by the `language` or `language_code` columns) are retained. In addition, the English and Hindi data points are translated into Bengali using Google Cloud Translation API. All columns from the original dataset are retained.

A handful of inaccuracies arising out of translation have been fixed so far. Therefore, the dataset can be a bit noisy. This is particularly true for coding related questions and answers. Moreover, some non-Bengali characters can be found in the text. In addition, potential duplicates from the original dataset are retained as well.