ulut / README.md
keramjan's picture
Upload dataset
b9d01c0 verified
metadata
dataset_info:
  - config_name: nug
    features:
      - name: question
        dtype: string
      - name: answer
        dtype: string
      - name: distractor1
        dtype: string
      - name: distractor2
        dtype: string
    splits:
      - name: train
        num_bytes: 7269
        num_examples: 31
    download_size: 6720
    dataset_size: 7269
  - config_name: wag
    features:
      - name: word
        dtype: string
      - name: antonym
        dtype: string
      - name: distractor1
        dtype: string
      - name: distractor2
        dtype: string
      - name: distractor3
        dtype: string
    splits:
      - name: train
        num_bytes: 3378
        num_examples: 50
    download_size: 4710
    dataset_size: 3378
  - config_name: wsm
    features:
      - name: index
        dtype: int64
      - name: word
        dtype: string
      - name: synonym
        dtype: string
      - name: distractor1
        dtype: string
      - name: distractor2
        dtype: string
      - name: distractor3
        dtype: string
    splits:
      - name: train
        num_bytes: 55784
        num_examples: 475
    download_size: 36639
    dataset_size: 55784
  - config_name: wub
    features:
      - name: statement
        dtype: string
      - name: check
        dtype: string
    splits:
      - name: train
        num_bytes: 2123
        num_examples: 28
    download_size: 2239
    dataset_size: 2123
  - config_name: wum
    features:
      - name: question
        dtype: string
      - name: answer
        dtype: string
      - name: A
        dtype: string
      - name: B
        dtype: string
      - name: C
        dtype: string
    splits:
      - name: train
        num_bytes: 3311
        num_examples: 22
    download_size: 4301
    dataset_size: 3311
configs:
  - config_name: nug
    data_files:
      - split: train
        path: nug/train-*
  - config_name: wag
    data_files:
      - split: train
        path: wag/train-*
  - config_name: wsm
    data_files:
      - split: train
        path: wsm/train-*
  - config_name: wub
    data_files:
      - split: train
        path: wub/train-*
  - config_name: wum
    data_files:
      - split: train
        path: wum/train-*
license: mit
task_categories:
  - text-generation
  - text-classification
  - question-answering
language:
  - ug
size_categories:
  - n<1K

Introduction

The ULUT (Uyghur language understanding test) dataset is aimed to evaluate LLM'm ability to understand Uyghur language. All the data are crowdsourced from open materials on the Internet. Currently there are 5 types of datasets: 1. wub 2. wum 3. nug 4. wsm 5. wag

1. WUB

This is a boolean type word understanding dataset. This dataset evaluates LLM's ability to use right word in right context.

2. WUM

This is a multiple-choice word understanding dataset that tests LLM's ability to use words correctly.

3. NUG

This is a generative dataset that tests LLM's ability to understand the basic natural events on Earth.

4. WSM

This is a multiple-choice dataset that tests LLM's ability to choose the synonym for a given word.

5. WAG

This is a generative dataset that tests LLM's ability to generate antonym for a given word.

Links

Evaluate your favorite LLM with this dataset using this Colab Notebook.