BERTTokenizer for C#

Source Code of NuGet package for tokenizing sentences and creating input for BERT Models.
· Report Bug · Request Feature

Table of Contents

1. About The Project
   • Built With
2. Getting Started
   • Prerequisites
   • Installation
3. Usage
4. License
5. Contact
6. Acknowledgments

About The Project

While working with BERT Models from Huggingface in combination with ML.NET, I stumbled upon several challenges. I documented them in here.
However, the biggest challenge by far was that I needed to implement my own tokenizer and pair them with the correct vocabulary. So, I decided to extend it and publish my implementation as a NuGet package and an open-source project. More info about this project can be found in this blog post.

This repository contains tokenizers for following models:
· BERT Base
· BERT Large
· BERT German
· BERT Multilingual
· BERT Base Uncased
· BERT Large Uncased

There are also clases using which you can upload your own vocabulary.

(back to top)

Built With

• .NET 6

(back to top)

Getting Started

The project is available as NuGet package.

Installation

To add BERT Tokenizers to your project use dotnet command:

dotnet add package BERTTokenizers

Or install it with package manager:

Install-Package BERTTokenizers

Usage

For example, you want to use Huggingface BERT Base Model whose input is defined like this:

public class BertInput
{
    [VectorType(1, 256)]
    [ColumnName("input_ids")]
    public long[] InputIds { get; set; }

    [VectorType(1, 256)]
    [ColumnName("attention_mask")]
    public long[] AttentionMask { get; set; }

    [VectorType(1, 256)]
    [ColumnName("token_type_ids")]
    public long[] TypeIds { get; set; }
}

For this you need to encode sentences like this:

var sentence = "I love you";

var tokenizer = new BertBaseTokenizer();

var encoded = tokenizer.Encode(256, sentence);

var bertInput = new BertInput()
                {
                    InputIds = encoded.Select(t => t.InputIds).ToArray(),
                    AttentionMask = encoded.Select(t => t.AttentionMask).ToArray(),
                    TypeIds = encoded.Select(t => t.TokenTypeIds).ToArray()
                };

For more examples, please refer to this Blog Post

See the open issues for a full list of proposed features (and known issues).

Contributing

Contributions are what make the open source community such an amazing place to learn, inspire, and create. Any contributions you make are greatly appreciated.

If you have a suggestion that would make this better, please fork the repo and create a pull request. You can also simply open an issue with the tag "enhancement". Don't forget to give the project a star! Thanks again!

1. Fork the Project
2. Create your Feature Branch (git checkout -b feature/AmazingFeature)
3. Commit your Changes (git commit -m 'Add some AmazingFeature')
4. Push to the Branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

(back to top)

License

Distributed under the MIT License. See LICENSE.txt for more information.

(back to top)

Contact

Nikola M. Zivkovic
[email protected]
LinkedIn
@NMZivkovic

(back to top)

Acknowledgments

• Gianluca Bertani - Performance Improvements
• Paul Calot - First Token bugfix

(back to top)