Extending Semantic Kernel using OllamaSharp for Chat and Text Completion

[elbruno]

Hi!

In previous posts I shared how to host and chat with a Llama 2 model hosted locally with Ollama. (view post). And then, it was time to learn how to integrate Semantic Kernel with OllamaSharp (nuget package and repo).

OllamaSharp is a .NET binding for the Ollama API, making it easy to interact with Ollama using your favorite .NET languages.

So, I decided to try it, and create a Chat Completion and a Text Generation specific implementation for Semantic Kernel using this library.

The full test is a console app using both services with Semantic Kernel.

Text Generation Service

The Text Generation Service is an easy one. Just implement the interface Microsoft.SemanticKernel.TextGeneration.ITextGenerationService, and the generated code looks like this:

//    Copyright (c) 2024
//    Author      : Bruno Capuano
//    Change Log  :
//    – Sample Text Generation Service for Ollama models
//
//    The MIT License (MIT)
//
//    Permission is hereby granted, free of charge, to any person obtaining a copy
//    of this software and associated documentation files (the "Software"), to deal
//    in the Software without restriction, including without limitation the rights
//    to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
//    copies of the Software, and to permit persons to whom the Software is
//    furnished to do so, subject to the following conditions:
//
//    The above copyright notice and this permission notice shall be included in
//    all copies or substantial portions of the Software.
//
//    THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
//    IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
//    FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
//    AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
//    LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
//    OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
//    THE SOFTWARE.

using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.ChatCompletion;
using OllamaSharp;
using OllamaSharp.Models.Chat;

namespace sk_ollamacsharp
{
    public class OllamaTextGenerationService : Microsoft.SemanticKernel.TextGeneration.ITextGenerationService
    {
        // public property for the model url endpoint
        public string ModelUrl { get; set; }
        public string ModelName { get; set; }

        public IReadOnlyDictionary<string, object?> Attributes => throw new NotImplementedException();

        public IAsyncEnumerable<StreamingTextContent> GetStreamingTextContentsAsync(string prompt, PromptExecutionSettings? executionSettings = null, Kernel? kernel = null, CancellationToken cancellationToken = default)
        {
            throw new NotImplementedException();
        }

        public async Task<IReadOnlyList<TextContent>> GetTextContentsAsync(string prompt, PromptExecutionSettings? executionSettings = null, Kernel? kernel = null, CancellationToken cancellationToken = default)
        {
            var ollama = new OllamaApiClient(ModelUrl, ModelName);

            var completionResponse = await ollama.GetCompletion(prompt, null, CancellationToken.None);

            TextContent stc = new TextContent(completionResponse.Response);
            return new List<TextContent> { stc };
        }
    }
}

Chat Completion Service

The chat completion, requires the implementation of the interface: IChatCompletionService. The code looks like this:

//    Copyright (c) 2024
//    Author      : Bruno Capuano
//    Change Log  :
//    – Sample Chat Completion Service for Ollama models
//
//    The MIT License (MIT)
//
//    Permission is hereby granted, free of charge, to any person obtaining a copy
//    of this software and associated documentation files (the "Software"), to deal
//    in the Software without restriction, including without limitation the rights
//    to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
//    copies of the Software, and to permit persons to whom the Software is
//    furnished to do so, subject to the following conditions:
//
//    The above copyright notice and this permission notice shall be included in
//    all copies or substantial portions of the Software.
//
//    THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
//    IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
//    FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
//    AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
//    LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
//    OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
//    THE SOFTWARE.

using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.ChatCompletion;
using OllamaSharp;
using OllamaSharp.Models.Chat;

namespace sk_ollamacsharp
{
    public class OllamaChatCompletionService : IChatCompletionService
    {
        // public property for the model url endpoint
        public string ModelUrl { get; set; }
        public string ModelName { get; set; }

        public IReadOnlyDictionary<string, object?> Attributes => throw new NotImplementedException();

        public async Task<IReadOnlyList<ChatMessageContent>> GetChatMessageContentsAsync(ChatHistory chatHistory, PromptExecutionSettings? executionSettings = null, Kernel? kernel = null, CancellationToken cancellationToken = default)
        {
            var ollama = new OllamaApiClient(ModelUrl, ModelName); // (uri);

            var chat = new Chat(ollama, _ => { });


            // iterate though chatHistory Messages
            foreach (var message in chatHistory)
            {
                if (message.Role == AuthorRole.System)
                {
                    await chat.SendAs(ChatRole.System, message.Content);
                    continue;
                }
            }

            var lastMessage = chatHistory.LastOrDefault();

            string question = lastMessage.Content;
            var chatResponse = "";
            var history = (await chat.Send(question, CancellationToken.None)).ToArray();

            var last = history.Last();
            chatResponse = last.Content;

            chatHistory.AddAssistantMessage(chatResponse);

            return chatHistory;
        }

        public IAsyncEnumerable<StreamingChatMessageContent> GetStreamingChatMessageContentsAsync(ChatHistory chatHistory, PromptExecutionSettings? executionSettings = null, Kernel? kernel = null, CancellationToken cancellationToken = default)
        {
            throw new NotImplementedException();
        }

    }
}

Test Chat Completion and Text Generation Services

With both services implemented, we can now code with Semantic Kernel to access these services.

The following code:

• Creates 2 services: text and chat, both with ollamasharp implementation
• Create a semantic kernel builder, register both services, and build a kernel.
• Using the kernel run a text generation sample, and later a chat history sample.
• In the chat sample, it also uses a System Message to define the chat behavior for the conversation.
• This is a test, there are a lot of improvements that can be made here.

//    Copyright (c) 2024
//    Author      : Bruno Capuano
//    Change Log  :
//    – Sample console application to use llama2 LLM running locally in Ubuntu with Semantic Kernel
//
//    The MIT License (MIT)
//
//    Permission is hereby granted, free of charge, to any person obtaining a copy
//    of this software and associated documentation files (the "Software"), to deal
//    in the Software without restriction, including without limitation the rights
//    to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
//    copies of the Software, and to permit persons to whom the Software is
//    furnished to do so, subject to the following conditions:
//
//    The above copyright notice and this permission notice shall be included in
//    all copies or substantial portions of the Software.
//
//    THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
//    IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
//    FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
//    AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
//    LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
//    OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
//    THE SOFTWARE.

using Microsoft.Extensions.DependencyInjection;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.ChatCompletion;
using Microsoft.SemanticKernel.TextGeneration;
using sk_ollamacsharp;

// llama2 in Ubuntu local in WSL
var ollamaChat = new OllamaChatCompletionService();
ollamaChat.ModelUrl = "http://localhost:11434";
ollamaChat.ModelName = "llama2";

var ollamaText = new OllamaTextGenerationService();
ollamaText.ModelUrl = "http://localhost:11434";
ollamaText.ModelName = "llama2";

// semantic kernel builder
var builder = Kernel.CreateBuilder();
builder.Services.AddKeyedSingleton<IChatCompletionService>("ollamaChat", ollamaChat);
builder.Services.AddKeyedSingleton<ITextGenerationService>("ollamaText", ollamaText);
var kernel = builder.Build();

// text generation
var textGen = kernel.GetRequiredService<ITextGenerationService>();
var response = textGen.GetTextContentsAsync("The weather in January in Toronto is usually ").Result;
Console.WriteLine(response[^1].Text);

// chat
var chat = kernel.GetRequiredService<IChatCompletionService>();
var history = new ChatHistory();
history.AddSystemMessage("You are a useful assistant that replies using a funny style and emojis. Your name is Goku.");
history.AddUserMessage("hi, who are you?");

// print response
var result = await chat.GetChatMessageContentsAsync(history);
Console.WriteLine(result[^1].Content);

The full code is available here: https://github.com/elbruno/semantickernel-localLLMs. And the main readme of the repo needs to be also updates.

Happy coding!

Greetings

Bruno Capuano