Add extension method GetStructuredResponse for IOpenAiClient

rodion-m · Jul 2, 2023 · c546c07 · c546c07
1 parent 16d08e8
commit c546c07
Show file tree

Hide file tree

Showing 16 changed files with 448 additions and 26 deletions.
diff --git a/OpenAI.ChatGpt.AspNetCore/ChatGPTFactory.cs b/OpenAI.ChatGpt.AspNetCore/ChatGPTFactory.cs
@@ -127,9 +127,9 @@ public async Task<ChatGPT> Create(
 
     public void Dispose()
     {
-        if (!_isHttpClientInjected)
+        if (!_isHttpClientInjected && _client is IDisposable disposableClient)
         {
-            _client.Dispose();
+            disposableClient.Dispose();
         }
     }
 }
diff --git a/OpenAI.ChatGpt.Modules.Translator/ChatGPTTranslatorService.cs b/OpenAI.ChatGpt.Modules.Translator/ChatGPTTranslatorService.cs
@@ -25,7 +25,7 @@ public ChatGPTTranslatorService(
 
     public ChatGPTTranslatorService(
         string apiKey, 
-        string? host,
+        string? host = null,
         string? defaultSourceLanguage = null,
         string? defaultTargetLanguage = null,
         string? extraPrompt = null)
@@ -39,9 +39,9 @@ public ChatGPTTranslatorService(
 
     public void Dispose()
     {
-        if (!_isHttpClientInjected)
+        if (!_isHttpClientInjected && _client is IDisposable disposableClient)
         {
-            _client.Dispose();
+            disposableClient.Dispose();
         }
     }
 
@@ -55,6 +55,14 @@ public async Task<string> Translate(
         if (text == null) throw new ArgumentNullException(nameof(text));
         var sourceLanguageOrDefault = sourceLanguage ?? _defaultSourceLanguage;
         var targetLanguageOrDefault = targetLanguage ?? _defaultTargetLanguage;
+        if (sourceLanguageOrDefault is null)
+        {
+            throw new ArgumentNullException(nameof(sourceLanguage), "Source language is not specified");
+        }
+        if (targetLanguageOrDefault is null)
+        {
+            throw new ArgumentNullException(nameof(targetLanguage), "Target language is not specified");
+        }
         var prompt = GetPrompt(sourceLanguageOrDefault, targetLanguageOrDefault);
         var response = await _client.GetChatCompletions(
             Dialog.StartAsSystem(prompt).ThenUser(text),

diff --git a/OpenAI.ChatGpt/ChatGPT.cs b/OpenAI.ChatGpt/ChatGPT.cs
@@ -92,9 +92,9 @@ public void Dispose()
     {
         Stop();
         _currentChat?.Dispose();
-        if (!_isClientInjected)
+        if (!_isClientInjected && _client is IDisposable disposableClient)
         {
-            _client.Dispose();
+            disposableClient.Dispose();
         }
     }
 

diff --git a/OpenAI.ChatGpt/ChatService.cs b/OpenAI.ChatGpt/ChatService.cs
@@ -93,13 +93,16 @@ private async Task<string> GetNextMessageResponse(
         cancellationToken = _cts.Token;
 
         var history = await LoadHistory(cancellationToken);
-        var messages = history.Append(message);
+        var messages = history.Append(message).ToArray();
 
         IsWriting = true;
         try
         {
+            var (model, maxTokens) = FindOptimalModelAndMaxToken(messages);
             var response = await _client.GetChatCompletionsRaw(
                 messages,
+                maxTokens: maxTokens,
+                model: model,
                 user: Topic.Config.PassUserIdToOpenAiRequests is true ? UserId : null,
                 requestModifier: Topic.Config.ModifyRequest,
                 cancellationToken: cancellationToken
@@ -117,7 +120,13 @@ await _chatHistoryStorage.SaveMessages(
             IsWriting = false; 
         }
     }
-
+
+    private (string model, int maxTokens) FindOptimalModelAndMaxToken(ChatCompletionMessage[] messages)
+    {
+        return ChatCompletionMessage.FindOptimalModelAndMaxToken(
+            messages, Topic.Config.Model, Topic.Config.MaxTokens);
+    }
+
     public IAsyncEnumerable<string> StreamNextMessageResponse(
         string message,
         bool throwOnCancellation = true,
@@ -143,11 +152,14 @@ private async IAsyncEnumerable<string> StreamNextMessageResponse(
         cancellationToken = _cts.Token;
 
         var history = await LoadHistory(cancellationToken);
-        var messages = history.Append(message);
+        var messages = history.Append(message).ToArray();
         var sb = new StringBuilder();
         IsWriting = true;
+        var (model, maxTokens) = FindOptimalModelAndMaxToken(messages);
         var stream = _client.StreamChatCompletions(
             messages,
+            maxTokens: maxTokens,
+            model: model,
             user: Topic.Config.PassUserIdToOpenAiRequests is true ? UserId : null,
             requestModifier: Topic.Config.ModifyRequest,
             cancellationToken: cancellationToken

diff --git a/OpenAI.ChatGpt/IOpenAiClient.cs b/OpenAI.ChatGpt/IOpenAiClient.cs
@@ -1,10 +1,9 @@
-using System.Runtime.CompilerServices;
-using OpenAI.ChatGpt.Models.ChatCompletion;
+using OpenAI.ChatGpt.Models.ChatCompletion;
 using OpenAI.ChatGpt.Models.ChatCompletion.Messaging;
 
 namespace OpenAI.ChatGpt;
 
-public interface IOpenAiClient : IDisposable
+public interface IOpenAiClient
 {
     Task<string> GetChatCompletions(
         UserOrSystemMessage dialog,
@@ -13,6 +12,7 @@ Task<string> GetChatCompletions(
         float temperature = ChatCompletionTemperatures.Default,
         string? user = null,
         Action<ChatCompletionRequest>? requestModifier = null,
+        Action<ChatCompletionResponse>? rawResponseGetter = null,
         CancellationToken cancellationToken = default);
 
     Task<string> GetChatCompletions(
@@ -22,6 +22,7 @@ Task<string> GetChatCompletions(
         float temperature = ChatCompletionTemperatures.Default,
         string? user = null,
         Action<ChatCompletionRequest>? requestModifier = null,
+        Action<ChatCompletionResponse>? rawResponseGetter = null,
         CancellationToken cancellationToken = default);
 
     Task<ChatCompletionResponse> GetChatCompletionsRaw(
@@ -81,8 +82,7 @@ IAsyncEnumerable<string> StreamChatCompletions(
         CancellationToken cancellationToken = default);
 
     IAsyncEnumerable<string> StreamChatCompletions(
-        ChatCompletionRequest request,
-        [EnumeratorCancellation] CancellationToken cancellationToken = default);
+        ChatCompletionRequest request,CancellationToken cancellationToken = default);
 
     IAsyncEnumerable<ChatCompletionResponse> StreamChatCompletionsRaw(
         ChatCompletionRequest request, CancellationToken cancellationToken = default);

diff --git a/OpenAI.ChatGpt/Models/ChatCompletion/ChatCompletionModels.cs b/OpenAI.ChatGpt/Models/ChatCompletion/ChatCompletionModels.cs
@@ -199,7 +199,7 @@ public static void EnsureMaxTokensIsSupported(string model, int maxTokens)
         {
             throw new ArgumentOutOfRangeException(
                 nameof(maxTokens), 
-                $"Max tokens must be less than or equal to {limit} for model {model}"
+                $"Max tokens must be less than or equal to {limit} for model {model} but was {maxTokens}"
             );
         }
     }
@@ -210,7 +210,7 @@ public static void EnsureMaxTokensIsSupportedByAnyModel(int maxTokens)
         if (maxTokens > limit)
         {
             throw new ArgumentOutOfRangeException(
-                nameof(maxTokens), $"Max tokens must be less than or equal to {limit}");
+                nameof(maxTokens), $"Max tokens must be less than or equal to {limit} but was {maxTokens}");
         }
     }
 }
diff --git a/OpenAI.ChatGpt/Models/ChatCompletion/Messaging/ChatCompletionMessage.cs b/OpenAI.ChatGpt/Models/ChatCompletion/Messaging/ChatCompletionMessage.cs
@@ -95,4 +95,34 @@ public override string ToString()
             ? $"{Role}: {Content}"
             :  string.Join(Environment.NewLine, _messages.Select(m => $"{m.Role}: {m.Content}"));
     }
+
+    public static (string model, int maxTokens) FindOptimalModelAndMaxToken(
+        IEnumerable<ChatCompletionMessage> messages,
+        string? model, 
+        int? maxTokens,
+        string smallModel = ChatCompletionModels.Default,
+        string bigModel = ChatCompletionModels.Gpt3_5_Turbo_16k,
+        bool useMaxPossibleTokens = true)
+    {
+        var tokenCount = CalculateApproxTotalTokenCount(messages);
+        switch (model, maxTokens)
+        {
+            case (null, null):
+            {
+                model = tokenCount > 6000 ? bigModel : smallModel;
+                maxTokens = GetMaxPossibleTokens(model);
+                break;
+            }
+            case (null, _):
+                model = smallModel;
+                break;
+            case (_, null):
+                maxTokens = useMaxPossibleTokens ? GetMaxPossibleTokens(model) : ChatCompletionRequest.MaxTokensDefault;
+                break;
+        }
+
+        return (model, maxTokens.Value);
+
+        int GetMaxPossibleTokens(string s) => ChatCompletionModels.GetMaxTokensLimitForModel(s) - tokenCount - 500;
+    }
 }
diff --git a/OpenAI.ChatGpt/OpenAIClient.cs → OpenAI.ChatGpt/OpenAiClient.cs b/OpenAI.ChatGpt/OpenAIClient.cs → OpenAI.ChatGpt/OpenAiClient.cs
@@ -12,7 +12,7 @@ namespace OpenAI.ChatGpt;
 
 /// <summary> Thread-safe OpenAI client. </summary>
 [Fody.ConfigureAwait(false)]
-public class OpenAiClient : IDisposable, IOpenAiClient
+public class OpenAiClient : IOpenAiClient, IDisposable
 {
     private const string DefaultHost = "https://api.openai.com/v1/";
     private const string ImagesEndpoint = "images/generations";
@@ -122,6 +122,7 @@ public async Task<string> GetChatCompletions(
         float temperature = ChatCompletionTemperatures.Default,
         string? user = null,
         Action<ChatCompletionRequest>? requestModifier = null,
+        Action<ChatCompletionResponse>? rawResponseGetter = null,
         CancellationToken cancellationToken = default)
     {
         if (dialog == null) throw new ArgumentNullException(nameof(dialog));
@@ -135,6 +136,7 @@ public async Task<string> GetChatCompletions(
             requestModifier
         );
         var response = await GetChatCompletionsRaw(request, cancellationToken);
+        rawResponseGetter?.Invoke(response);
         return response.Choices[0].Message!.Content;
     }
 
@@ -145,6 +147,7 @@ public async Task<string> GetChatCompletions(
         float temperature = ChatCompletionTemperatures.Default,
         string? user = null,
         Action<ChatCompletionRequest>? requestModifier = null,
+        Action<ChatCompletionResponse>? rawResponseGetter = null,
         CancellationToken cancellationToken = default)
     {
         if (messages == null) throw new ArgumentNullException(nameof(messages));
@@ -158,6 +161,7 @@ public async Task<string> GetChatCompletions(
             requestModifier
         );
         var response = await GetChatCompletionsRaw(request, cancellationToken);
+        rawResponseGetter?.Invoke(response);
         return response.GetMessageContent();
     }