El Bruno

Hi!

In previous posts I shared how to host and chat with a Llama 2 model hosted locally with Ollama. (view post).

And then I also found 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 };
	}
	}
	}

view raw ollamasharpsktextgeneration.cs hosted with ❤ by GitHub

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();
	}
	}
	}

view raw ollamasharpskchatcompletion.cs hosted with ❤ by GitHub

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);

view raw ollamasharpskchattexttest.cs hosted with ❤ by GitHub

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