AI Integration?

[VictorioBerra]

LinqPad recently added AI integration and its super helpful. Any plans for NetPad?

[tareqimbasher]

It is planned. I haven't figured out how it would work exactly. I'm open to any suggestions, do you like the way its implemented in LINQPad?

[VictorioBerra]

I love the shift+space which triggers the generation. The personal assistant implementation is a little odd since the UI is built into the result view area, try it and you will see. I think VSCode does that a lot better.

[tareqimbasher]

Thank you for the feedback!

[tareqimbasher]

There is #127 which is related, asking for GitHub Copilot. LINQPad uses ChatGPT for code generation. I'd love to hear what everyone's preferences are.

[VictorioBerra]

LinqPad can use azure and openai. Not sure if ollama is planned.

Older screenshot from last year ^

[tareqimbasher]

Ollama would be pretty neat too.

[frankhaugen]

Ollama would be pretty neat too.

LM Studio expose Ollama and other local models as OpenAI -equvalent.

[frankhaugen]

Shouldn't be too hard to lift the code completion example from Microsoft.Extensions.Ai, or Semantic Kernel and plop it into the UI. I believe it might be worth looking at the Semantic Memory samples aswell. With LM Studio, which makes it trivial to run LLM code on local.

[tareqimbasher]

Thanks @frankhaugen. If you can provide links that would be helpful.

[frankhaugen]

Thanks @frankhaugen. If you can provide links that would be helpful.

So I did a script:

I'm using LM studio locally to host and run the LLM

var modelId = "meta-llama-3.1-8b-instruct";
var endpoint = new Uri("http://localhost:1234/v1");
var kernel = Kernel.CreateBuilder()
    .AddOpenAIChatCompletion(
        modelId: modelId,
        apiKey: null,
        endpoint: endpoint)
    .Build();

OpenAIChatCompletionService chatCompletionService = new(modelId: modelId, apiKey: null, endpoint: endpoint);

var systemPrompt = """
You are an expert C# copilot. Continue the user’s function using clean, idiomatic code.
Return only valid C# that compiles. Avoid repeats.
You only return code
""";

var userPrefix = """
static int Fibonacci(int n)
{
""";

var userPrompt = $"""
Continue the function implementation:
{userPrefix}
""";

var chatHistory = new ChatHistory();
chatHistory.AddSystemMessage(systemPrompt);
chatHistory.AddUserMessage(userPrompt);
chatHistory.AddAssistantMessage("Sure, here is the continuation of the Fibonacci function:");

await foreach (var word in chatCompletionService.GetStreamingChatMessageContentsAsync(
    chatHistory, new PromptExecutionSettings() {
    }, kernel))
{
    Console.Write($"{word} ");
}

Output, (it needs tweaks and parsing, which is what the copilot plugin in VS Code mainly does):

 

 ``` c sharp 
 static  int  Fibonacci (int  n )
 {
      if  ( n  <=   1 )  return  n ;
      else  return  Fibonacci (n  -   1 )  +  Fibonacci (n  -   2 );
 }
 `` `

 However ,  this  recursive  approach  has  a  high  time  complexity  and  may  cause  stack  overflow  for  large  inputs .  A  more  efficient  solution  would  be  to  use  memo ization :

 ``` c sharp 
 static  int  Fibonacci (int  n )
 {
      if  ( n  <=   1 )  return  n ;
      int []  fib  =  new  int [n  +   1 ];
      fib [ 0 ]  =   0 ;  
      fib [ 1 ]  =   1 ;

      for  ( int  i  =   2 ;  i  <=  n ;  i ++)
          fib [i ]  =  fib [i  -   1 ]  +  fib [i  -   2 ];

      return  fib [n ];
 }
 `` `

 Or  use  a  simple  iterative  approach :

 ``` c sharp 
 static  int  Fibonacci (int  n )
 {
      if  ( n  <=   1 )  return  n ;
      int  a  =   0 ,  b  =   1 ;

      for  ( int  i  =   2 ;  i  <=  n ;  i ++)
          a  =  b ;
          b  =  a  +  b ;

      return  b ;
 }