TextBox 2.0 is a comprehensive Python library designed for text generation tasks, offering a unified and standardized framework for applying pre-trained language models (PLMs). It caters to researchers and developers working with various text generation applications, providing a reproducible and flexible environment for experimentation.

How It Works

TextBox 2.0 standardizes the pipeline for PLM-based text generation, supporting 13 common tasks and 83 datasets. It integrates 47 diverse PLMs, categorized by function (e.g., translation, dialogue, controllable generation), and offers 4 pre-training objectives and efficient training strategies like distributed data parallel. This unified approach simplifies the application of state-of-the-art models and facilitates result reproduction.

Quick Start & Requirements

• Install via install.sh after cloning the repository. A conda environment with Python 3.8 is recommended.
• Datasets need to be downloaded from Hugging Face and placed in a dataset folder.
• Example run command: python run_textbox.py --model=BART --dataset=samsum --model_path=facebook/bart-base
• Official documentation for models and datasets is available.

Highlighted Details

• Supports 13 text generation tasks and 83 datasets.
• Integrates 47 pre-trained language models/modules.
• Offers 4 pre-training objectives and efficient training strategies.
• Includes 17 automatic evaluation metrics and visualization tools.

Maintenance & Community

Developed and maintained by AI Box. Contributions are welcomed via issues and pull requests.

Licensing & Compatibility

MIT License. Permissive for commercial use and closed-source linking.

Limitations & Caveats

The installation process may require specific handling for dependencies like files2rouge. Users need to manually download datasets from Hugging Face.