README.md

---
dataset_info:
  features:
  - name: question
    dtype: string
  - name: subject
    dtype: string
  - name: choices
    sequence: string
  - name: answer
    dtype: int64
  - name: task
    dtype: string
  splits:
  - name: train
    num_bytes: 162899630
    num_examples: 99842
  download_size: 47653197
  dataset_size: 162899630
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
license: apache-2.0
task_categories:
- text-generation
language:
- en
pretty_name: MMLU auxiliary trained set labelled by e5 mistral 7b instruct
size_categories:
- 10K<n<100K
---

# Dataset Card for MMLU Auxiliary Trained Set Labelled by e5-mistral-7b-instruct

## Table of Contents
- [Dataset Description](#dataset-description)
  - [Dataset Summary](#dataset-summary)
  - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
  - [Languages](#languages)
- [Dataset Structure](#dataset-structure)
  - [Data Instances](#data-instances)
  - [Data Fields](#data-fields)
  - [Data Splits](#data-splits)
- [Dataset Creation](#dataset-creation)
  - [Curation Rationale](#curation-rationale)
  - [Source Data](#source-data)
  - [Annotations](#annotations)
  - [Personal and Sensitive Information](#personal-and-sensitive-information)
- [Considerations for Using the Data](#considerations-for-using-the-data)
  - [Social Impact of Dataset](#social-impact-of-dataset)
  - [Discussion of Biases](#discussion-of-biases)
  - [Other Known Limitations](#other-known-limitations)
- [Additional Information](#additional-information)
  - [Dataset Curators](#dataset-curators)
  - [Licensing Information](#licensing-information)
  - [Citation Information](#citation-information)
  - [Contact Information](#contact-information)
  - [Visualizations](#visualizations)

## Dataset Description

### Dataset Summary
This dataset, named "MMLU Auxiliary Trained Set Labelled by e5-mistral-7b-instruct," consists of 99,842 examples spanning various subjects. Each instance includes a question, multiple choice options, a subject category, and an answer. The unique aspect of this dataset is the task label for each question, generated by a zero-shot classifier constructed from the `intfloat/e5-mistral-7b-instruct` model and trained on the auxiliary set of the Massive Multitask Language Understanding (MMLU).

### Supported Tasks and Leaderboards
This dataset supports text-generation tasks. It is particularly useful for training and evaluating models on a wide range of subjects using the task labels generated by the zero-shot classifier.

### Languages
The dataset is predominantly in English.

## Dataset Structure

### Data Instances
A typical data instance in this dataset comprises:
- `question`: A textual question or prompt.
- `subject`: The subject category of the question.
- `choices`: A list of possible answers.
- `answer`: The correct answer's index from the choices.
- `task`: The task label assigned by the zero-shot classifier.

### Data Fields
- `question`: string
- `subject`: string
- `choices`: sequence of strings
- `answer`: int64
- `task`: string

### Data Splits
- Train Split: 99,842 examples

## Dataset Creation

### Curation Rationale
The dataset was curated to enhance the diversity and scope of language models in understanding and generating responses across a wide range of subjects. The use of a zero-shot classifier for task labelling introduces a novel approach to categorizing and understanding textual data.

### Source Data
The data was sourced from the auxiliary-train set of MMLU and processed to include task labels generated by the `intfloat/e5-mistral-7b-instruct` model.

### Annotations
#### Annotation process
The task labels were generated automatically by a zero-shot classifier model, specifically `intfloat/e5-mistral-7b-instruct`.
#### Who are the annotators?
There were no human annotators; the process was entirely automated using the zero-shot classifier.

### Personal and Sensitive Information
The dataset does not contain personal or sensitive information as it is focused on general knowledge questions and subjects.

## Considerations for Using the Data

### Social Impact of Dataset
This dataset can aid in developing more versatile and knowledgeable language models, potentially impacting various domains like education, research, and AI development.

### Discussion of Biases
Given the automated nature of task label generation and diverse subject matter, biases may be minimal but could still exist based on the underlying training data of the zero-shot classifier.

### Other Known Limitations
The primary limitation is the reliance on the zero-shot classifier's accuracy for task labeling, which may not always align with human judgment.

## Additional Information

### Dataset Curators
The dataset was curated by the team involved in the development of `mmlu`.

### Licensing Information
The dataset is available under the Apache-2.0 License.

### Citation Information
@misc{mmlu_auxiliary_trained_set,
  title         = {{MMLU Auxiliary Trained Set Labelled by e5-mistral-7b-instruct}},
  author        = {Kaizhao Liang},
  year          = {2024},
  howpublished  = {https://huggingface.co/datasets/kz919/mmlu-auxiliary-train-e5-mistral-7b-instruct},
  note          = {Accessed: Date of Access},
  description   = {A dataset of 99,842 examples across various subjects, each including a question, multiple choice options, a subject category, an answer, and a task label generated by a zero-shot classifier constructed from the intfloat/e5-mistral-7b-instruct model.},
  license       = {Apache-2.0}
}

### Contact Information
[homepage](https://kyleliang919.github.io/)

### Visualizations
- **Counts by Category**
  <img src="https://huggingface.co/datasets/kz919/mmlu-auxiliary-train-e5-mistral-7b-instruct/resolve/main/per_category_distribution.png">

- **Counts by Super Category**
  <img src="https://huggingface.co/datasets/kz919/mmlu-auxiliary-train-e5-mistral-7b-instruct/resolve/main/super_category_distribution.png">

---