bartpho word

Introduction
BARTpho is a pre-trained sequence-to-sequence model specifically designed for the Vietnamese language. It includes two versions: BARTpho-syllable and BARTpho-word. These models are based on the architecture of BART, a denoising model optimized for generative NLP tasks. BARTpho has demonstrated superior performance in Vietnamese text summarization compared to the mBART model, setting a new state-of-the-art.

Architecture
BARTpho utilizes the "large" architecture of the BART model, which is designed for sequence-to-sequence tasks. It is particularly effective for generative tasks due to its pre-training scheme focused on denoising.

Training
The training of BARTpho involved a comprehensive pre-training phase using large-scale Vietnamese text data. This process allows the model to perform exceptionally well in tasks like Vietnamese text summarization, as evidenced by both automatic and human evaluations.

Guide: Running Locally

1. Install the necessary libraries: Ensure that PyTorch and the Hugging Face Transformers library are installed.
2. Clone the repository: Download the BARTpho model from its GitHub repository.
3. Load the model: Use the Hugging Face Transformers library to load the BARTpho model.
4. Run inference: Input Vietnamese text data to generate summaries or perform other NLP tasks.

For optimal performance, consider using cloud GPU services like AWS, Google Cloud, or Azure.

License
BARTpho is licensed under the MIT License, which allows for a wide range of uses, including commercial applications, as long as the appropriate credit is given.