responses: token counting

Author: lofcz

src/LlmTornado/Chat/ConversationCompression.cs

@@ -112,7 +112,7 @@ public async Task<List<ChatMessage>> Compress(List<ChatMessage> messages, string
         Task<string>[] summaryTasks = chunks.Select(chunk => SummarizeChunk(api, chunk, options, context, token)).ToArray();
         string[] summaries = await Task.WhenAll(summaryTasks);
 
-        List<ChatMessage> conversation = new List<ChatMessage>();
+        List<ChatMessage> conversation = [];
 
         foreach (string summary in summaries)
         {
@@ -203,8 +203,8 @@ private async Task<string> SummarizeChunk(TornadoApi api, List<ChatMessage> chun
 
 public class ConversationContent
 {
-    public List<ChatMessage> SystemMessages { get; set; } = new List<ChatMessage>();
-    public List<ChatMessage> MessagesToCompress { get; set; } = new List<ChatMessage>();
+    public List<ChatMessage> SystemMessages { get; set; } = [];
+    public List<ChatMessage> MessagesToCompress { get; set; } = [];
 
     /// <summary>
     ///  Get the content of the conversation based on the compression options

src/LlmTornado/Chat/Vendors/Cohere/VendorCohereChatRequest.cs

@@ -94,7 +94,7 @@ public VendorCohereChatRequest(ChatRequest request, IEndpointProvider provider)
         if (request.Tools is { Count: > 0 })
         {
             // Optimize: single loop instead of Where().Select().ToList() to avoid intermediate enumerable allocations
-            List<VendorCohereTool> tools = new List<VendorCohereTool>();
+            List<VendorCohereTool> tools = [];
             foreach (var t in request.Tools)
             {
                 if (t.Function is not null)

src/LlmTornado/Chat/Vendors/Mistral/MistralChatMessageConverter.cs

@@ -32,7 +32,7 @@ internal class MistralChatMessageConverter : JsonConverter<ChatMessage>
         {
             StringBuilder reasoningBuilder = new StringBuilder();
             StringBuilder textBuilder = new StringBuilder();
-            List<ChatMessagePart> parts = new List<ChatMessagePart>();
+            List<ChatMessagePart> parts = [];
 
             foreach (JToken block in contentArr)
             {
@@ -170,7 +170,7 @@ private static ChatMessageRoles ParseRole(string? role)
             return null;
         }
 
-        List<ToolCall> toolCalls = new List<ToolCall>();
+        List<ToolCall> toolCalls = [];
 
         foreach (JToken tc in toolCallsArray)
         {

src/LlmTornado/Chat/Vendors/Zai/VendorZaiChatRequestData.cs

@@ -82,14 +82,14 @@ public VendorZaiChatRequestData(ChatRequest request) : base(request)
         // Add built-in tools from vendor extensions
         if (request.VendorExtensions?.Zai?.BuiltInTools is { Count: > 0 })
         {
-            Tools ??= new List<VendorZaiTool>();
+            Tools ??= [];
             Tools.AddRange(request.VendorExtensions.Zai.BuiltInTools.Select(ConvertBuiltInTool));
         }
     }
 
     private static List<VendorZaiTool> ConvertTools(List<Tool> tools)
     {
-        List<VendorZaiTool> zaiTools = new();
+        List<VendorZaiTool> zaiTools = [];
 
         foreach (Tool tool in tools)
         {

src/LlmTornado/Code/DiffMatchPatch/PatchList.cs

@@ -187,7 +187,7 @@ private static bool IsHeaderlessHeader(string line)
 
     private static Patch? ParseHeaderlessPatch(IReadOnlyList<string> lines, ref int index)
     {
-        List<string> body = new();
+        List<string> body = [];
 
         while (index < lines.Count)
         {

src/LlmTornado/Infra/Clr.cs

@@ -67,7 +67,7 @@ internal static class Clr
 
     private static MethodInfo? GetCachedMethod(Type type, string methodName, BindingFlags bindingFlags, Type[]? parameterTypes)
     {
-        Type[] paramTypes = parameterTypes ?? Array.Empty<Type>();
+        Type[] paramTypes = parameterTypes ?? [];
         return methodInfoCache.GetOrAdd((type, methodName, paramTypes), key => 
             key.Type.GetMethod(key.MethodName, bindingFlags, null, key.ParameterTypes, null));
     }

src/LlmTornado/Ocr/Models/Mistral/OcrModelMistral.cs

@@ -17,7 +17,7 @@ public class OcrModelMistral : BaseVendorModelProvider
     /// OCR 3 (mistral-ocr-2512) - Released December 2024.
     /// Features: table_format (markdown/html), extract_header, extract_footer, hyperlinks output.
     /// </summary>
-    public static readonly OcrModel ModelOcr2512 = new OcrModel("mistral-ocr-2512", LLmProviders.Mistral, new List<string> { "mistral-ocr-latest" });
+    public static readonly OcrModel ModelOcr2512 = new OcrModel("mistral-ocr-2512", LLmProviders.Mistral, ["mistral-ocr-latest"]);
 
     /// <summary>
     /// <inheritdoc cref="ModelOcr2512"/>

src/LlmTornado/Responses/ResponseInputTokensResult.cs

@@ -0,0 +1,22 @@
+using Newtonsoft.Json;
+
+namespace LlmTornado.Responses;
+
+/// <summary>
+/// Result of the input token counting endpoint (POST /responses/input_tokens).
+/// Returns the number of input tokens the request would consume.
+/// </summary>
+public class ResponseInputTokensResult
+{
+    /// <summary>
+    /// The object type. Always "response.input_tokens".
+    /// </summary>
+    [JsonProperty("object")]
+    public string Object { get; set; } = "response.input_tokens";
+
+    /// <summary>
+    /// The number of input tokens the request would use.
+    /// </summary>
+    [JsonProperty("input_tokens")]
+    public int InputTokens { get; set; }
+}

src/LlmTornado/Responses/ResponsesEndpoint.cs

@@ -284,6 +284,35 @@ public async Task<HttpCallResult<ResponseCompactResult>> CompactResponseSafe(Res
         return await HttpPost<ResponseCompactResult>(provider, Endpoint, url: GetUrl(provider, "/compact"), postData: body, ct: request.CancellationToken).ConfigureAwait(false);
     }
 
+    /// <summary>
+    /// Returns the number of input tokens a response request would consume, without actually generating a response.
+    /// </summary>
+    /// <param name="request">The request to count tokens for.</param>
+    /// <param name="cancellationToken">Cancellation token.</param>
+    public async Task<ResponseInputTokensResult> CountInputTokens(ResponseRequest request, CancellationToken cancellationToken = default)
+    {
+        HttpCallResult<ResponseInputTokensResult> data = await CountInputTokensSafe(request, cancellationToken).ConfigureAwait(false);
+
+        if (!data.Ok)
+        {
+            throw data.Exception;
+        }
+
+        return data.Data;
+    }
+
+    /// <summary>
+    /// Returns the number of input tokens a response request would consume, without actually generating a response.
+    /// </summary>
+    /// <param name="request">The request to count tokens for.</param>
+    /// <param name="cancellationToken">Cancellation token.</param>
+    public async Task<HttpCallResult<ResponseInputTokensResult>> CountInputTokensSafe(ResponseRequest request, CancellationToken cancellationToken = default)
+    {
+        IEndpointProvider provider = Api.GetProvider(request.Model ?? ChatModel.OpenAi.Gpt35.Turbo);
+        TornadoRequestContent requestBody = request.Serialize(provider);
+        return await HttpPost<ResponseInputTokensResult>(provider, Endpoint, url: GetUrl(provider, "/input_tokens"), postData: requestBody.Body, model: request.Model, ct: cancellationToken).ConfigureAwait(false);
+    }
+
     /// <summary>
     /// Creates a responses API request.
     /// </summary>

src/LlmTornado/Skills/CreateSkillRequest.cs

@@ -30,7 +30,7 @@ public class CreateSkillRequest
     /// </summary>
     public CreateSkillRequest()
     {
-        Files = Array.Empty<FileUploadRequest>();
+        Files = [];
     }
 
     /// <summary>
@@ -41,7 +41,7 @@ public CreateSkillRequest()
     public CreateSkillRequest(string displayTitle, FileUploadRequest[]? files = null)
     {
         DisplayTitle = displayTitle;
-        Files = files ?? Array.Empty<FileUploadRequest>();
+        Files = files ?? [];
     }
 
     /// <summary>