ZhEnOnnx 1.0.0

ZhEnOnnx

ZhEnOnnx is a Windows .NET translation library for the supplied Marian ONNX models. It supports Chinese-to-English and English-to-Chinese translation, preferring DirectML GPU inference and automatically falling back to CPU when DirectML cannot initialize.

Add a project reference

<ItemGroup>
  <ProjectReference Include="..\ZhEnOnnx\ZhEnOnnx.csproj" />
</ItemGroup>

Use the translator

The models\zh-en and models\en-zh folders are part of this library. They are copied next to your application automatically, so no model path is required.

using ZhEnOnnx;

using var translator = new BilingualTranslator();

string english = translator.TranslateToEnglish("你好，世界！");
string chinese = translator.TranslateToChinese("Machine learning is useful.");

Console.WriteLine(translator.Backend); // DirectMl or Cpu

To force CPU inference, use new BilingualTranslator(preferGpu: false). You can also provide a custom model directory with new BilingualTranslator(modelRoot); it must contain the zh-en and en-zh subdirectories.

Publish as a NuGet package

The package includes both models (about 856 MB total). Create a package with:

dotnet pack -c Release --no-restore

When another .NET project installs that package, its build output and published output receive a models folder automatically. The parameterless constructor then works unchanged.

Improve English-to-Chinese output

English-to-Chinese calls split multi-sentence input, normalize Chinese punctuation and remove unwanted whitespace by default. Use Terminology for deterministic replacements of model output.

var options = new TranslationOptions
{
    Terminology = new Dictionary<string, string>
    {
        ["机器 学习"] = "机器学习",
        ["ONNX运行时"] = "ONNX Runtime"
    }
};

string chinese = translator.TranslateToChinese("Machine learning is useful.", options);

Translate C# identifiers

PascalCase, camelCase and underscore identifiers are expanded into normal words before English-to-Chinese translation. For example, BilingualTranslator becomes Bilingual Translator and _service becomes service, so they can be translated semantically.

var options = new TranslationOptions
{
    ExpandCodeIdentifiers = true
};

string chinese = translator.TranslateToChinese(
    "BilingualTranslator uses _service to configure service.",
    options);

Set PreserveCodeIdentifiers = true when identifiers must remain unchanged in translated output.

Return a C# identifier from Chinese-to-English translation

Use EnglishOutputIdentifierStyle to remove spaces and punctuation from an English translation.

var options = new TranslationOptions
{
    EnglishOutputIdentifierStyle = EnglishIdentifierStyle.CamelCase
};

string identifier = translator.TranslateToEnglish("双语翻译器", options);
// Example result: bilingualTranslator