|
--- |
|
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. |
