Tokenizers.DotNet

• Tokenizers.DotNet
• Nuget Package list
• Requirements
• Supported functionalities
• How to use
  • (1) Install the packages
  • (2) Write the code
• How to build

Tokenizers.DotNet

.NET wrapper of HuggingFace Tokenizers library

Nuget Package list

Package	main	Description
Tokenizers.DotNet		Core library
Tokenizers.DotNet.runtime.win		Native bindings for windows x64

Requirements

• .NET 6 or above

Supported functionalities

• Download tokenizer files from Hugginface Hub
• Load tokenizer file(.json) from local
• Decode embeddings to string

How to use

(1) Install the packages

1. From the NuGet, install Tokenizers.DotNet package
2. And then, install Tokenizers.DotNet.runtime.win package too

(2) Write the code

Check following example code:

using Tokenizers.DotNet;

// Download skt/kogpt2-base-v2/tokenizer.json from the hub
var hubName = "skt/kogpt2-base-v2";
var filePath = "tokenizer.json";
var fileFullPath = await HuggingFace.GetFileFromHub(hubName, filePath, "deps");
Console.WriteLine($"Downloaded {fileFullPath}");

// Create a tokenizer instance
var tokenizer = new Tokenizer(vocabPath: fileFullPath);
var tokens = new uint[] { 9330, 387, 12857, 9376, 18649, 9098, 7656, 6969, 8084, 1 };
var decoded = tokenizer.Decode(tokens);
Console.WriteLine($"Decoded: {decoded}");

Console.WriteLine($"Version of Tokenizers.DotNet.runtime.win: {tokenizer.GetVersion()}");

How to build

1. Prepare following stuff:
   1. Rust build system (cargo)
   2. .NET build system (dotnet 6.0)
   3. PowerShell (Recommend 7.4.2 or above)
2. Run build_all_clean.ps1
   1. To build Tokenizers.DotNet.runtime.win only, run build_rust.ps1
   2. To build Tokenizers.DotNet only, run build_dotnet.ps1

Each build artifacts will be in nuget directory.