Semantic-Fleet is a repository designed to extend the capabilities of Semantic Kernel. It focuses on providing connectors to small large language models (e.g. Llamas), and to provide tools for distributing work to a fleet of models, with ChatGPT serving as the captain of the fleet. This repository is more than just a collection of existing connectors; it's a platform for future innovations in the .NET ecosystem for AI.

We would like to inform our users that due to recent changes in the Oobabooga API (see commit 454fcf3 from 13/11/2023), any versions of Oobabooga beyond this commit will no longer be supported by semantic-fleet.

The designer of Oobabooga has replaced the traditional API with a new one modeled after OpenAI's API. Unfortunately, we have not yet had the opportunity to update our bridge to be compatible with these changes.

We are actively working to ensure compatibility in future releases, but for now, we recommend our users to:

1. Avoid updating Oobabooga beyond the specified commit if you wish to continue using semantic-fleet without interruption.
2. Stay tuned for our future updates for the support of the new Oobabooga API.

We appreciate your understanding and patience as we work through these changes.

A robust connector that currently covers the main Oobabooga's specific blocking and streaming completion and chat APIs.

📖 Learn More:

• Installing Oobabooga and Configuring Multi-Start Scripts
• Oobabooga Connector Guide
• Don't forget to check-out notebooks. They provide a great overview of what's possible with our published connectors.

Install the package via NuGet:

dotnet add package MyIA.SemanticKernel.Connectors.AI.Oobabooga

In .Net interactive :

#r "nuget: MyIA.SemanticKernel.Connectors.AI.Oobabooga"

Different settings are used for text and chat completion, both in blocking and streaming modes. Here's a quick example for text completion:

var settings = new OobaboogaTextCompletionSettings(endpoint: new Uri("https://localhost/"),  blockingPort: 5000, streamingPort: 5005);
var oobabooga = new OobaboogaTextCompletion(settings);

// Get text completions
var completions = await oobabooga.GetCompletionsAsync("Hello, world!", new OobaboogaCompletionRequestSettings());

Why stick to one when you can have many? MultiConnector lets you integrate multiple LLMs seamlessly, optimizing for speed and cost. It intelligently offloads tasks from a primary, more expensive connector to a secondary, more cost-effective one without sacrificing reliability nor performance.

📖 Learn More: MultiConnector Guide

Install the package via NuGet:

dotnet add package MyIA.SemanticKernel.Connectors.AI.MultiConnector

In .Net interactive :

#r "nuget: MyIA.SemanticKernel.Connectors.AI.MultiConnector

Multiconnector has many settings controlling how to route text completion calls, and how to autmatically sample completions from a primary connector, test, evaluate and update the routing settings to use secondary connectors.

var settings = new MultiTextCompletionSettings();

// (...) Creating a primary openAiNamedCompletion and secondary  oobaboogaCompletions

var builder = Microsoft.SemanticKernel.Kernel.Builder;

builder.WithMultiConnectorCompletionService(
    serviceId: null,
    settings: settings,
    mainTextCompletion: openAiNamedCompletion,
    setAsDefault: true,
    analysisTaskCancellationToken: cleanupToken.Token,
    otherCompletions: oobaboogaCompletions.ToArray());

var kernel = builder.Build();

// Get text completion from primary connector first
 var result = await kernel.RunAsync(semanticFunctionOrPlan, contextVariables, cancellationToken: cleanupToken.Token).ConfigureAwait(false);

 // (...) Peform analysis manually or automatically depending on settings

// Get text completion from secondary connectors
 var optimizedResult = await kernel.RunAsync(semanticFunctionOrPlan, contextVariables, cancellationToken: cleanupToken.Token).ConfigureAwait(false);

For a detailed overview of how to fill the gaps, please refer to the notebooks and integration tests.

Want a overview of what's possible with our published connectors? Our .Net interactive notebooks are a great place to start.

📖 Learn More: Notebooks Guide

• Open AI API: Oobabooga offers a dedicated extension mimicking Open AI API. It extends support to embeddings and image generation models. This will be available as a separate package.
• Probabilistic MultiConnector: We will be adding some Infer.Net magic to make MultiConnector even smarter. More specifically, the following examples will be merged and integrated into the model vetting process.
  • Student Skills
  • Assessing People's Skills
  • Difficulty vs Ability
  • Calibrating reviews
• Spark.Net Integration: Get ready to host a cluster of mini local LLMs.

We'll be providing NuGet packages for both the Oobabooga Connector and MultiConnector for easier integration into your projects.

Here is the Nuget Package for Oobabooga connector

And here is the Nuget Package for Multiconnector

Got something to add? We'd love to see it. Check out our contributing guidelines.

Got something you'd like to get added? Do you want those future features already? We'd love you to get in touch !