---

ONNX Chat Streaming in C#

This page was automatically generated by AI; not yet reviewed for accuracy...

The content and code samples on this page were generated by using the ai CLI with customized prompts in this repository.

It's cool, but, it's experimental. 😁

Please review the content and code before using it in your application.

This sample demonstrates how to use the ONNX Chat API with streaming in a C# console application.

Program.cs
OnnxGenAIChatCompletionsStreamingClass.cs
get-phi3-mini-onnx.cmd
Phi3ChatStreaming.csproj

How to generate this sample

Command

ai dev new phi3-onnx-chat-streaming --csharp

Output

AI - Azure AI CLI, Version 1.0.0
Copyright (c) 2024 Microsoft Corporation. All Rights Reserved.

This PUBLIC PREVIEW version may change at any time.
See: https://aka.ms/azure-ai-cli-public-preview

Generating 'phi3-onnx-chat-streaming' in 'phi3-onnx-chat-streaming-cs' (4 files)...

OnnxGenAIChatCompletionsStreamingClass.cs
Program.cs
get-phi3-mini-onnx.cmd
Phi3ChatStreaming.csproj

Generating 'phi3-onnx-chat-streaming' in 'phi3-onnx-chat-streaming-cs' (4 files)... DONE!

Program.cs

STEP 1: Read the configuration settings from environment variables:

Program.cs

var modelDirectory = Environment.GetEnvironmentVariable("ONNX_GENAI_MODEL_PATH") ?? "<insert your ONNX GenAI model path here>";
var systemPrompt = Environment.GetEnvironmentVariable("ONNX_GENAI_SYSTEM_PROMPT") ?? "You are a helpful assistant.";

STEP 2: Initialize the helper class with the configuration settings:

Program.cs

var chat = new OnnxGenAIChatStreamingClass(modelDirectory, systemPrompt);

STEP 3: Obtain user input, use the helper class to get the assistant's response, and display responses as they are received:

Program.cs

while (true)
{
    Console.Write("User: ");
    var input = Console.ReadLine();
    if (string.IsNullOrEmpty(input) || input == "exit") break;

    Console.Write("\nAssistant: ");
    chat.GetChatCompletionStreaming(input, update => {
        Console.Write(update);
    });
    Console.WriteLine("\n");
}

OnnxGenAIChatCompletionsStreamingClass.cs

STEP 1: Create the client and initialize chat message history with a system message:

OnnxGenAIChatCompletionsStreamingClass.cs

public OnnxGenAIChatStreamingClass(string modelDirectory, string systemPrompt)
{
    _modelDirectory = modelDirectory;
    _systemPrompt = systemPrompt;

    _messages = new List<ContentMessage>();
    _messages.Add(new ContentMessage { Role = "system", Content = systemPrompt });

    _model = new Model(modelDirectory);
    _tokenizer = new Tokenizer(_model);
}

STEP 2: When the user provides input, add the user message to the chat message history:

OnnxGenAIChatCompletionsStreamingClass.cs

public string GetChatCompletionStreaming(string userPrompt, Action<string>? callback = null)
{
    _messages.Add(new ContentMessage { Role = "user", Content = userPrompt });

STEP 3: Send the chat message history to the streaming ONNX Chat API and process each update:

OnnxGenAIChatCompletionsStreamingClass.cs

    var responseContent = string.Empty;
    using var tokens = _tokenizer.Encode(string.Join("\n", _messages
        .Select(m => $"<|{m.Role}|>\n{m.Content}\n<|end|>"))
        + "<|assistant|>\n");

    using var generatorParams = new GeneratorParams(_model);
    generatorParams.SetSearchOption("max_length", 2048);
    generatorParams.SetInputSequences(tokens);

    using var generator = new Generator(_model, generatorParams);

    var sb = new StringBuilder();
    while (!generator.IsDone())
    {
        generator.ComputeLogits();
        generator.GenerateNextToken();

        var outputTokens = generator.GetSequence(0);
        var newToken = outputTokens.Slice(outputTokens.Length - 1, 1);

        var output = _tokenizer.Decode(newToken);
        sb.Append(output);

        callback?.Invoke(output);
    }

    responseContent = sb.ToString();
    _messages.Add(new ContentMessage { Role = "assistant", Content = responseContent });

    return responseContent;
}