SemanticKernel.Connectors.OnnxRuntimeGenAI

Semantic Kernel connector for ONNX models.

How to use

Prerequisites

ONNX models, for example Phi-3 Mini-4K-Instruct

git lfs install
git clone https://huggingface.co/microsoft/Phi-3-mini-4k-instruct-onnx

Code

Create a new console app and add Nuget Package:

-- for CPU
feiyun0112.SemanticKernel.Connectors.OnnxRuntimeGenAI.CPU

-- for CUDA
feiyun0112.SemanticKernel.Connectors.OnnxRuntimeGenAI.CUDA

Then change Program.cs to:

Kernel kernel = Kernel.CreateBuilder()
           .AddOnnxRuntimeGenAIChatCompletion(
               modelPath: @"d:\Phi-3-mini-4k-instruct-onnx\cpu_and_mobile\cpu-int4-rtn-block-32-acc-level-4")
           .Build();

string prompt = @"Write a joke";

await foreach (string text in kernel.InvokePromptStreamingAsync<string>(prompt,
                   new KernelArguments(new OnnxRuntimeGenAIPromptExecutionSettings() { MaxLength = 2048 })))
{
    Console.Write(text);
}

使用说明

先决条件

你需要下载所需的ONNX模型，例如 `Phi-3 Mini-4K-Instruct

git lfs install
git clone https://hf-mirror.com/microsoft/Phi-3-mini-4k-instruct-onnx

示例代码

创建新的控制台应用，并根据你的硬件配置选择合适的Nuget包:

-- for CPU
feiyun0112.SemanticKernel.Connectors.OnnxRuntimeGenAI.CPU

-- for CUDA
feiyun0112.SemanticKernel.Connectors.OnnxRuntimeGenAI.CUDA

然后，只需几行代码，你就可以构建Kernel，并开始生成聊天内容:

Kernel kernel = Kernel.CreateBuilder()
           .AddOnnxRuntimeGenAIChatCompletion(
               modelPath: @"d:\Phi-3-mini-4k-instruct-onnx\cpu_and_mobile\cpu-int4-rtn-block-32-acc-level-4")
           .Build();

string prompt = @"Write a joke";

await foreach (string text in kernel.InvokePromptStreamingAsync<string>(prompt,
                   new KernelArguments(new OnnxRuntimeGenAIPromptExecutionSettings() { MaxLength = 2048 })))
{
    Console.Write(text);
}