Merge branch 'master' into release

# Conflicts: # src/Directory.Build.props
rodion-m · Dec 17, 2023 · 99152b4 · 99152b4
2 parents fa9c23c + 613d5f6
commit 99152b4
Show file tree

Hide file tree

Showing 12 changed files with 82 additions and 97 deletions.
diff --git a/src/Directory.Build.props b/src/Directory.Build.props
@@ -1,6 +1,6 @@
 <Project>
     <PropertyGroup>
-        <Version>4.0.2-alpha</Version>
+        <Version>4.1.0-alpha</Version>
         <TargetFrameworks>net6.0;net7.0;net8.0</TargetFrameworks>
         <ImplicitUsings>enable</ImplicitUsings>
         <Nullable>enable</Nullable>

diff --git a/src/OpenAI.ChatGpt.AspNetCore/AiClientFromConfiguration.cs b/src/OpenAI.ChatGpt.AspNetCore/AiClientFromConfiguration.cs
@@ -47,7 +47,7 @@ private static void ThrowUnkownProviderException(string provider)
     /// <inheritdoc />
     public Task<string> GetChatCompletions(
         UserOrSystemMessage dialog,
-        int maxTokens = ChatCompletionRequest.MaxTokensDefault,
+        int? maxTokens = null,
         string model = ChatCompletionModels.Default, float temperature = ChatCompletionTemperatures.Default,
         string? user = null, bool jsonMode = false, long? seed = null,
         Action<ChatCompletionRequest>? requestModifier = null,
@@ -60,7 +60,7 @@ public Task<string> GetChatCompletions(
     /// <inheritdoc />
     public Task<string> GetChatCompletions(
         IEnumerable<ChatCompletionMessage> messages,
-        int maxTokens = ChatCompletionRequest.MaxTokensDefault,
+        int? maxTokens = null,
         string model = ChatCompletionModels.Default, float temperature = ChatCompletionTemperatures.Default,
         string? user = null, bool jsonMode = false, long? seed = null,
         Action<ChatCompletionRequest>? requestModifier = null,
@@ -73,7 +73,7 @@ public Task<string> GetChatCompletions(
     /// <inheritdoc />
     public Task<ChatCompletionResponse> GetChatCompletionsRaw(
         IEnumerable<ChatCompletionMessage> messages,
-        int maxTokens = ChatCompletionRequest.MaxTokensDefault,
+        int? maxTokens = null,
         string model = ChatCompletionModels.Default, float temperature = ChatCompletionTemperatures.Default,
         string? user = null, bool jsonMode = false, long? seed = null,
         Action<ChatCompletionRequest>? requestModifier = null,
@@ -86,7 +86,7 @@ public Task<ChatCompletionResponse> GetChatCompletionsRaw(
     /// <inheritdoc />
     public IAsyncEnumerable<string> StreamChatCompletions(
         IEnumerable<ChatCompletionMessage> messages,
-        int maxTokens = ChatCompletionRequest.MaxTokensDefault,
+        int? maxTokens = null,
         string model = ChatCompletionModels.Default, float temperature = ChatCompletionTemperatures.Default,
         string? user = null, bool jsonMode = false, long? seed = null,
         Action<ChatCompletionRequest>? requestModifier = null,
@@ -99,7 +99,7 @@ public IAsyncEnumerable<string> StreamChatCompletions(
     /// <inheritdoc />
     public IAsyncEnumerable<string> StreamChatCompletions(
         UserOrSystemMessage messages,
-        int maxTokens = ChatCompletionRequest.MaxTokensDefault, string model = ChatCompletionModels.Default,
+        int? maxTokens = null, string model = ChatCompletionModels.Default,
         float temperature = ChatCompletionTemperatures.Default, string? user = null, bool jsonMode = false,
         long? seed = null, Action<ChatCompletionRequest>? requestModifier = null,
         CancellationToken cancellationToken = default)
@@ -108,6 +108,12 @@ public IAsyncEnumerable<string> StreamChatCompletions(
             messages, maxTokens, model, temperature, user, jsonMode, seed, requestModifier, cancellationToken);
     }
 
+    /// <inheritdoc />
+    public int? GetDefaultMaxTokens(string model)
+    {
+        return _client.GetDefaultMaxTokens(model);
+    }
+
     /// <inheritdoc />
     public IAsyncEnumerable<string> StreamChatCompletions(
         ChatCompletionRequest request,
@@ -124,5 +130,10 @@ public IAsyncEnumerable<ChatCompletionResponse> StreamChatCompletionsRaw(
         return _client.StreamChatCompletionsRaw(request, cancellationToken);
     }
 
+    public string GetOptimalModel(ChatCompletionMessage[] messages)
+    {
+        return _client.GetOptimalModel(messages);
+    }
+
     internal IAiClient GetInnerClient() => _client;
 }
diff --git a/src/OpenAI.ChatGpt/ChatService.cs b/src/OpenAI.ChatGpt/ChatService.cs
@@ -102,11 +102,10 @@ private async Task<string> GetNextMessageResponse(
         IsWriting = true;
         try
         {
-            var (model, maxTokens) = FindOptimalModelAndMaxToken(messages);
             var response = await _client.GetChatCompletionsRaw(
                 messages,
-                maxTokens: maxTokens,
-                model: model,
+                maxTokens: Topic.Config.MaxTokens,
+                model:Topic.Config.Model ?? _client.GetOptimalModel(message),
                 user: Topic.Config.PassUserIdToOpenAiRequests is true ? UserId : null,
                 requestModifier: Topic.Config.ModifyRequest,
                 cancellationToken: cancellationToken
@@ -125,12 +124,6 @@ await _chatHistoryStorage.SaveMessages(
         }
     }
 
-    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,
@@ -159,11 +152,10 @@ private async IAsyncEnumerable<string> StreamNextMessageResponse(
         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,
+            maxTokens: Topic.Config.MaxTokens,
+            model:Topic.Config.Model ?? _client.GetOptimalModel(message),
             user: Topic.Config.PassUserIdToOpenAiRequests is true ? UserId : null,
             requestModifier: Topic.Config.ModifyRequest,
             cancellationToken: cancellationToken

diff --git a/src/OpenAI.ChatGpt/IAiClient.cs b/src/OpenAI.ChatGpt/IAiClient.cs
@@ -8,6 +8,17 @@ namespace OpenAI.ChatGpt;
 /// </summary>
 public interface IAiClient
 {
+    /// <summary>
+    /// Retrieves the default maximum number of tokens for a given model.
+    /// </summary>
+    /// <param name="model">
+    /// The model name for which to retrieve the maximum number of tokens.
+    /// </param>
+    /// <returns>
+    /// The default maximum number of tokens as an integer or just null if it's reqired to delegate it to the AI service.
+    /// </returns>
+    int? GetDefaultMaxTokens(string model);
+
     /// <summary>
     /// Get a chat completion response as a string
     /// </summary>
@@ -41,7 +52,7 @@ public interface IAiClient
     /// <returns>The chat completion response as a string</returns>
     Task<string> GetChatCompletions(
         UserOrSystemMessage dialog,
-        int maxTokens = ChatCompletionRequest.MaxTokensDefault,
+        int? maxTokens = null,
         string model = ChatCompletionModels.Default,
         float temperature = ChatCompletionTemperatures.Default,
         string? user = null,
@@ -84,7 +95,7 @@ Task<string> GetChatCompletions(
     /// <returns>The chat completion response as a string</returns>
     Task<string> GetChatCompletions(
         IEnumerable<ChatCompletionMessage> messages,
-        int maxTokens = ChatCompletionRequest.MaxTokensDefault,
+        int? maxTokens = null,
         string model = ChatCompletionModels.Default,
         float temperature = ChatCompletionTemperatures.Default,
         string? user = null,
@@ -126,7 +137,7 @@ Task<string> GetChatCompletions(
     /// <returns>The raw chat completion response</returns>
     Task<ChatCompletionResponse> GetChatCompletionsRaw(
         IEnumerable<ChatCompletionMessage> messages,
-        int maxTokens = ChatCompletionRequest.MaxTokensDefault,
+        int? maxTokens = null,
         string model = ChatCompletionModels.Default,
         float temperature = ChatCompletionTemperatures.Default,
         string? user = null,
@@ -167,7 +178,7 @@ Task<ChatCompletionResponse> GetChatCompletionsRaw(
     /// <returns>Chunks of LLM's response, one by one.</returns>
     IAsyncEnumerable<string> StreamChatCompletions(
         IEnumerable<ChatCompletionMessage> messages,
-        int maxTokens = ChatCompletionRequest.MaxTokensDefault,
+        int? maxTokens = null,
         string model = ChatCompletionModels.Default,
         float temperature = ChatCompletionTemperatures.Default,
         string? user = null,
@@ -201,7 +212,7 @@ IAsyncEnumerable<string> StreamChatCompletions(
     /// <returns>Chunks of LLM's response, one by one</returns>
     IAsyncEnumerable<string> StreamChatCompletions(
         UserOrSystemMessage messages,
-        int maxTokens = ChatCompletionRequest.MaxTokensDefault,
+        int? maxTokens = null,
         string model = ChatCompletionModels.Default,
         float temperature = ChatCompletionTemperatures.Default,
         string? user = null,
@@ -227,4 +238,7 @@ IAsyncEnumerable<string> StreamChatCompletions(
     /// <returns>A stream of raw chat completion responses</returns>
     IAsyncEnumerable<ChatCompletionResponse> StreamChatCompletionsRaw(
         ChatCompletionRequest request, CancellationToken cancellationToken = default);
+
+    string GetOptimalModel(ChatCompletionMessage[] messages);
+    string GetOptimalModel(UserOrSystemMessage dialog) => GetOptimalModel(dialog.GetMessages().ToArray());
 }
diff --git a/src/OpenAI.ChatGpt/Models/ChatCompletion/ChatCompletionRequest.cs b/src/OpenAI.ChatGpt/Models/ChatCompletion/ChatCompletionRequest.cs
@@ -14,9 +14,6 @@ namespace OpenAI.ChatGpt.Models.ChatCompletion;
 /// </remarks>
 public class ChatCompletionRequest
 {
-    public const int MaxTokensDefault = 64;
-
-    private int _maxTokens = MaxTokensDefault;
     private string _model = ChatCompletionModels.Default;
     private float _temperature = ChatCompletionTemperatures.Default;
     private IEnumerable<ChatCompletionMessage> _messages;
@@ -87,7 +84,6 @@ public float Temperature
 
     /// <summary>
     /// The maximum number of tokens allowed for the generated answer.
-    /// Defaults to <see cref="MaxTokensDefault"/>.
     /// This value is validated and limited with <see cref="ChatCompletionModels.GetMaxTokensLimitForModel"/> method.
     /// It's possible to calculate approximately tokens count using <see cref="ChatCompletionMessage.CalculateApproxTotalTokenCount()"/> method.
     /// </summary>
@@ -98,15 +94,7 @@ public float Temperature
     /// Encoding algorithm can be found here: https://github.com/latitudegames/GPT-3-Encoder
     /// </remarks>
     [JsonPropertyName("max_tokens")]
-    public int MaxTokens
-    {
-        get => _maxTokens;
-        set
-        {
-            ChatCompletionModels.EnsureMaxTokensIsSupported(Model, value);
-            _maxTokens = value;
-        }
-    }
+    public int? MaxTokens { get; set; } = null;
 
     /// <summary>
     /// Number between -2.0 and 2.0.

diff --git a/src/OpenAI.ChatGpt/Models/ChatCompletion/Messaging/ChatCompletionMessage.cs b/src/OpenAI.ChatGpt/Models/ChatCompletion/Messaging/ChatCompletionMessage.cs
@@ -113,34 +113,4 @@ 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/src/OpenAI.ChatGpt/Models/ChatGPTConfig.cs b/src/OpenAI.ChatGpt/Models/ChatGPTConfig.cs
@@ -56,7 +56,6 @@ public class ChatGPTConfig
 
     /// <summary>
     /// The maximum number of tokens allowed for the generated answer.
-    /// Defaults to <see cref="ChatCompletionRequest.MaxTokensDefault"/>.
     /// This value is validated and limited with <see cref="ChatCompletionModels.GetMaxTokensLimitForModel"/> method.
     /// It's possible to calculate approximately tokens count using <see cref="ChatCompletionMessage.CalculateApproxTotalTokenCount()"/> method.
     /// Maps to: <see cref="ChatCompletionRequest.MaxTokens"/>

diff --git a/src/OpenAI.ChatGpt/OpenAiClient.cs b/src/OpenAI.ChatGpt/OpenAiClient.cs
@@ -152,7 +152,7 @@ private static void ValidateHttpClient(
     /// <inheritdoc />
     public async Task<string> GetChatCompletions(
         UserOrSystemMessage dialog,
-        int maxTokens = ChatCompletionRequest.MaxTokensDefault,
+        int? maxTokens = null,
         string model = ChatCompletionModels.Default,
         float temperature = ChatCompletionTemperatures.Default,
         string? user = null,
@@ -185,7 +185,7 @@ public async Task<string> GetChatCompletions(
     /// <inheritdoc />
     public async Task<string> GetChatCompletions(
         IEnumerable<ChatCompletionMessage> messages,
-        int maxTokens = ChatCompletionRequest.MaxTokensDefault,
+        int? maxTokens = null,
         string model = ChatCompletionModels.Default,
         float temperature = ChatCompletionTemperatures.Default,
         string? user = null,
@@ -218,7 +218,7 @@ public async Task<string> GetChatCompletions(
     /// <inheritdoc />
     public async Task<ChatCompletionResponse> GetChatCompletionsRaw(
         IEnumerable<ChatCompletionMessage> messages,
-        int maxTokens = ChatCompletionRequest.MaxTokensDefault,
+        int? maxTokens = null,
         string model = ChatCompletionModels.Default,
         float temperature = ChatCompletionTemperatures.Default,
         string? user = null,
@@ -277,7 +277,7 @@ protected virtual string GetChatCompletionsEndpoint()
     /// <inheritdoc />
     public IAsyncEnumerable<string> StreamChatCompletions(
         IEnumerable<ChatCompletionMessage> messages,
-        int maxTokens = ChatCompletionRequest.MaxTokensDefault,
+        int? maxTokens = null,
         string model = ChatCompletionModels.Default,
         float temperature = ChatCompletionTemperatures.Default,
         string? user = null,
@@ -304,9 +304,9 @@ public IAsyncEnumerable<string> StreamChatCompletions(
         return StreamChatCompletions(request, cancellationToken);
     }
 
-    private static ChatCompletionRequest CreateChatCompletionRequest(
+    private ChatCompletionRequest CreateChatCompletionRequest(
         IEnumerable<ChatCompletionMessage> messages,
-        int maxTokens,
+        int? maxTokens,
         string model,
         float temperature,
         string? user,
@@ -316,6 +316,7 @@ private static ChatCompletionRequest CreateChatCompletionRequest(
         Action<ChatCompletionRequest>? requestModifier)
     {
         ArgumentNullException.ThrowIfNull(messages);
+        maxTokens ??= GetDefaultMaxTokens(model);
         var request = new ChatCompletionRequest(messages)
         {
             Model = model,
@@ -330,10 +331,15 @@ private static ChatCompletionRequest CreateChatCompletionRequest(
         return request;
     }
 
+    public int? GetDefaultMaxTokens(string model)
+    {
+        return null;
+    }
+
     /// <inheritdoc />
     public IAsyncEnumerable<string> StreamChatCompletions(
         UserOrSystemMessage messages,
-        int maxTokens = ChatCompletionRequest.MaxTokensDefault,
+        int? maxTokens = null,
         string model = ChatCompletionModels.Default,
         float temperature = ChatCompletionTemperatures.Default,
         string? user = null,
@@ -346,7 +352,8 @@ public IAsyncEnumerable<string> StreamChatCompletions(
         if (model == null) throw new ArgumentNullException(nameof(model));
         EnsureJsonModeIsSupported(model, jsonMode);
         ThrowIfDisposed();
-        var request = CreateChatCompletionRequest(messages.GetMessages(),
+        var request = CreateChatCompletionRequest(
+            messages.GetMessages(),
             maxTokens,
             model,
             temperature,
@@ -393,7 +400,12 @@ public IAsyncEnumerable<ChatCompletionResponse> StreamChatCompletionsRaw(
             cancellationToken
         );
     }
-
+
+    public string GetOptimalModel(ChatCompletionMessage[] messages)
+    {
+        return ChatCompletionModels.Gpt4Turbo;
+    }
+
     private static void EnsureJsonModeIsSupported(string model, bool jsonMode)
     {
         if(jsonMode && !ChatCompletionModels.IsJsonModeSupported(model))

diff --git a/src/internal/OpenAI.GeneratedKiotaClient/GeneratedClientsFactory.cs b/src/internal/OpenAI.GeneratedKiotaClient/GeneratedClientsFactory.cs
@@ -11,7 +11,8 @@ public static GeneratedOpenAiClient CreateGeneratedOpenAiClient(HttpClient httpC
         ArgumentNullException.ThrowIfNull(httpClient);
         var authProvider = new AnonymousAuthenticationProvider();
         var adapter = new HttpClientRequestAdapter(authProvider, httpClient: httpClient);
-        return new GeneratedOpenAiClient(adapter);
+        var openAiClient = new GeneratedOpenAiClient(adapter);
+        return openAiClient;
     }
 
     public static GeneratedAzureOpenAiClient CreateGeneratedAzureOpenAiClient(HttpClient httpClient)

diff --git a/...OpenAI.ChatGpt.Modules.StructuredResponse/OpenAiClientExtensions.GetStructuredResponse.cs b/...OpenAI.ChatGpt.Modules.StructuredResponse/OpenAiClientExtensions.GetStructuredResponse.cs
@@ -114,17 +114,11 @@ internal static async Task<TObject> GetStructuredResponse<TObject>(
         {
             editMsg.Content += GetAdditionalJsonResponsePrompt(responseFormat, examples, jsonSerializerOptions);
 
-            (model, maxTokens) = FindOptimalModelAndMaxToken(
-                dialog.GetMessages(), 
-                model, 
-                maxTokens,
-                smallModel: ChatCompletionModels.Gpt4,
-                bigModel: ChatCompletionModels.Gpt4
-            );
+            model ??= client.GetOptimalModel(dialog);
 
             var response = await client.GetChatCompletions(
                 dialog,
-                maxTokens.Value,
+                maxTokens,
                 model,
                 temperature,
                 user,