---
dataset_info:
  features:
  - name: id
    dtype: string
  - name: text
    dtype: string
  - name: lang
    dtype: string
  - name: type
    dtype: string
  splits:
  - name: train
    num_bytes: 10228099668
    num_examples: 2573755
  download_size: 5810967149
  dataset_size: 10228099668
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
license: odc-by
language:
- fr
- en
- es
tags:
- C/C++
- Python
- JavaScript
- Java
size_categories:
- 1M<n<10M
---

# Dataset-Tokenizer

This dataset is designed for training models to classify both textual and code data. It provides a diverse collection of natural language and code samples, aimed at tasks such as natural language processing (NLP) and code understanding. The dataset supports three natural languages (English, French, Spanish) and five programming languages (Python, Java, JavaScript, C++, C).

## Languages and Sources

- **Natural Languages** (80%):
  - English (34%)
  - French (33%)
  - Spanish (33%)

- **Programming Languages** (20%):
  - Python (25%)
  - Java (25%)
  - JavaScript (25%)
  - C++ (12.5%)
  - C (12.5%)

- **Data Types**:
  - **NL**: Textual Content
  - **CL**: Code Snippets

## Dataset Features

Each entry in the dataset includes the following attributes:

- **id** (string): A unique identifier for each sample, represented as a SHA256 hash of the original text.
- **text** (string): The original content of the text or code snippet.
- **lang** (string): The language of the data, which can be either a natural language (English, French, Spanish) or a programming language (Python, Java, JavaScript, C++, C).
- **type** (string): The type of content. Use "NL" for natural language text and "CL" for code language (programming languages).

## Intended Use

This dataset is ideal for training machine learning models to perform tasks such as:

- Text classification and sentiment analysis in multiple natural languages.
- Code classification, code completion, or other code-related tasks in various programming languages.
- Cross-language or multi-language classification for both text and code.

It supports the development of models that can handle both human languages and programming languages, making it versatile for a wide range of NLP and code understanding tasks.

## Limitations

- The dataset may not cover all edge cases for each language or code type.
- The representation of languages may not be balanced; some languages may have more samples than others.
- Code snippets may vary in complexity and may not always represent typical coding practices.

## Citation

Please cite or link back to this dataset on Hugging Face Hub if used in your projects.