Codeblaze.SemanticKernel.Connectors.Ollama 1.3.1

Ollama Connector

Supports

• text generation
• chat completion
• embedding generation

⚠️ Embedding generation: Is experimental in the semantic kernel.

Quick Start

• Install from nuget

  dotnet add package Codeblaze.SemanticKernel.Connectors.Ollama
  

• Text Generation

  Configure the kernel

  var builder = Kernel.CreateBuilder();
  
  // provide the HTTP client used to interact with Ollama API
  builder.Services.AddTransient<HttpClient>();
  
  builder.AddOllamaTextGeneration(
   config["Ollama:Model"], // Ollama model Id
   config["Ollama:BaseUrlGeneration"] // Ollama endpoint
  );
  
  var kernel = builder.Build();
  

  Usage

  const string prompt = """
  Bot: How can I help you?
  User: {{$input}}
  
  ---------------------------------------------
  
  The intent of the user in 5 words or less:
  """;
  
  var result = await kernel.InvokePromptAsync(prompt, new KernelArguments
  {
   {"input", input}
  });
  
  System.Console.WriteLine(result.GetValue<string>());
  

• Chat Completion

  Configure the kernel

  var builder = new KernelBuilder();
  
  // provide the HTTP client used to interact with Ollama API
  builder.Services.AddTransient<HttpClient>();
  
  builder.AddOllamaChatCompletion(
   config["Ollama:Model"], // Ollama model Id
   config["Ollama:BaseUrlGeneration"] // Ollama endpoint
  );
  
  var kernel = builder.Build();
  

  Usage

  var chat = _Kernel.GetRequiredService<IChatCompletionService>();
  
  var history = new ChatHistory();
  
  // add messages to current chat history as required
  history.AddSystemMessage("...");
  history.AddAssistantMessage("...");
  history.AddUserMessage(input);
  
  // result is a list of all chat messages 
  // including the output of current prompt
  var result = await chat.GetChatMessageContentsAsync(history);
  
  // Print the last message
  System.Console.WriteLine(result[^1].Content);
  

• Embedding Generation (Experimental)

  Configure the kernel

  var builder = new KernelBuilder();
  
  // provide the HTTP client used to interact with Ollama API
  builder.Services.AddTransient<HttpClient>();
  
  builder.AddOllamaTextEmbeddingGeneration(
   config["Ollama:Model"], // Ollama model Id
   config["Ollama:BaseUrlGeneration"] // Ollama endpoint
  );
  
  // Configure memory backend (e.g Azure Cognitive Search)
  
  var kernel = builder.Build();
  

  Usage

  var memory = _Kernel.GetRequiredService<ISemanticTextMemory>();
  
  // This will internally call Ollama embedding service to generate embeddings
  memory.SaveReferenceAsync(
   collection: "collection",
   externalSourceName: "ext-collection",
   externalId: id, // reference id (database entity id)
   description: input,
   text: input
  );