fredericaltorres / fAI

AI Experimentation With Different Providers

fAI

Introduction

• AI Experimentation With Different Providers in C#
• OpenAI
  • Querying GPT
  • Image generation (DALL-E)
  • Embeddings
  • Speech To Text
  • Text To Speech
• Microsoft Cognitive Services
  • Text To Speech
  • Azure Search Index with Embeddings
• Deepgram
  • Speech To Text

Overview

Library fAI available on NuGet.

Querying OpenAI GPT models

Querying and expecting answer in JSON format.

[Fact()]
public void Completion_JsonMode_WorldCup()
{
    var client = new OpenAI();
    var p = new Prompt_GPT_35_Turbo_JsonAnswer
    {
        Messages = new List<GPTMessage>()
        {
            new GPTMessage { Role =  MessageRole.system, Content = "You are a helpful assistant designed to output JSON." },
            new GPTMessage { Role =  MessageRole.user,   Content = "Who won the soccer world cup in 1998?" }
        }
    };
    var response = client.Completions.Create(p);
    Assert.True(response.Success);
    var answer = response.JsonObject["winner"];
    Assert.Equal("France", answer);
}

Conversation and data analysis with GPT 4

[Fact()]
public void Completion_Chat_QuestionAboutPastSchedule()
{
    var client = new OpenAI();
    var prompt = new Prompt_GPT_4
    {
        Messages = new List<GPTMessage> 
        {
            new GPTMessage { Role =  MessageRole.system, Content = "You are a helpful assistant." },
            new GPTMessage { Role =  MessageRole.user,   Content = $"08/02/2021 15:00 Meeting with Eric." },
            new GPTMessage { Role =  MessageRole.user,   Content = $"09/01/2021 15:00 Meeting with Eric." },
            new GPTMessage { Role =  MessageRole.user,   Content = $"09/10/2021 10:00 Take the dog to the vet." },
            new GPTMessage { Role =  MessageRole.user,   Content = $"09/20/2021 15:00 Meeting with Rick and John" },
        }
    };
    var response = client.Completions.Create(prompt);

    prompt.Messages.Add(new GPTMessage { Role = MessageRole.user, Content = "When was the last time I talked with Eric?" });
    response = client.Completions.Create(prompt);
    Assert.Matches(@"Eric.*09\/01\/2021 at 15:00", response.Text);

    prompt.Messages.Add(new GPTMessage { Role = MessageRole.user, Content = "What do I have to do on 09/10/2021?" });
    response = client.Completions.Create(prompt);
    Assert.Matches(@"dog.*vet.*10:00", response.Text);
}

Translation with model gpt-3.5-turbo

public string Translate(string text, TranslationLanguages sourceLangague, TranslationLanguages targetLanguage);

const string ReferenceEnglishSentence = "Hello world.";
[Fact()]
public void Translate_EnglishToSpanish()
{
    var client = new OpenAI();
    var translation = client.Completions.Translate(ReferenceEnglishSentence, TranslationLanguages.English, TranslationLanguages.Spanish);
    Assert.Equal("'Hola mundo.'", translation);
}

Others helper to execute translations. Translating dictionary or list of strings can do 2 things

1. Translate more than one text with one call to GPT.
2. For a dictionary the key is not translated as part of the answer and can be use to map the translation with the original texts.

public Dictionary<string, string> Translate(Dictionary<string, string> inputDictionary, TranslationLanguages sourceLangague, TranslationLanguages targetLanguage);
public List<string>               Translate(List<string> strings, TranslationLanguages sourceLangague, TranslationLanguages targetLanguage);

Querying OpenAI Embeddings API

 public class OpenAIEmbeddingsTests
 {
     [Fact()]
     public void Embeddings_Create()
     {
         var input = "I am he as you are he as you are me. And we are all together. See how they run like pigs from a gun. See how they fly. I'm crying.";
         var client = new OpenAI();
         var r = client.Embeddings.Create(input);

         // r.Data[0].Embedding contains the list of float 

         Assert.Equal("embedding", r.Data[0].Object);
         Assert.Equal(r.Data[0].EmbeddingMaxValue, r.Data[0].Embedding.Count);
         Assert.Equal(37, r.Usage.PromptTokens);
         Assert.Equal(37, r.Usage.TotalTokens);
     }
 }

Querying OpenAI DALL-E. Image Generation

[Fact()]
public void Image_Generate()
{
    var prompt = @"Generate an image inspired by Victor Hugo's classic novel, 'Les Misérables'. 
The image should depict three characters, each with distinct characteristics. 
The first is an older, physically strong man with a scarred face, wearing threadbare clothes, indicative of a hard life — this represents Jean Valjean. 
The second is a young woman radiating innocence and kindness; she wears modest clothes and has beautiful shining eyes — this is Cosette. 
The third is a stern-looking middle-aged man in a gentleman's attire and hat,  representing law and order — representative of Javert. 
Their expressions should reflect the nuances of the complex relationships 
they share in the story.
";
    var client = new OpenAI();
    var r = client.Image.Generate(prompt, size :  OpenAIImageSize._1792x1024);
    var pngFileNames = r.DownloadImage();
    Assert.True(pngFileNames.Count == 1);
    Assert.True(File.Exists(pngFileNames[0]));
}

Third parties API Key

API keys can be hard coded in code or read automatically from environment variable.

Environment Variable Names
MICROSOFT_AZURE_SEARCH_KEY
MICROSOFT_COGNITIVE_SERVICES_KEY
MICROSOFT_COGNITIVE_SERVICES_REGION
OPENAI_API_KEY
OPENAI_ORGANIZATION_ID
OPENAI_LOG_FILE
DEEPGRAM_API_KEY