From d161a14b7c9335e6f2d9826c472bdf589c70ba71 Mon Sep 17 00:00:00 2001
From: Martin Evans <martindevans@gmail.com>
Date: Wed, 14 Oct 2026 23:35:46 +0100
Subject: [PATCH 1/5] - Updated nuget packages  - Updated all projects to
 dotnet 10 from dotnet 8.0  - Removed ImageSharp dependency

---
 LLama.Benchmark/LLama.Benchmark.csproj        |  2 +-
 .../Examples/MtmdInteractiveModeExecute.cs    | 25 ++++---------------
 LLama.Examples/LLama.Examples.csproj          | 16 ++++++------
 LLama.Unittest/LLama.Unittest.csproj          | 10 ++++----
 LLama.Web/LLama.Web.csproj                    |  4 +--
 LLama.WebAPI/LLama.WebAPI.csproj              |  2 +-
 LLama/LLamaSharp.csproj                       | 18 ++++++-------
 7 files changed, 30 insertions(+), 47 deletions(-)

diff --git a/LLama.Benchmark/LLama.Benchmark.csproj b/LLama.Benchmark/LLama.Benchmark.csproj
index 8d18e0251..17cbfbf20 100644
--- a/LLama.Benchmark/LLama.Benchmark.csproj
+++ b/LLama.Benchmark/LLama.Benchmark.csproj
@@ -3,7 +3,7 @@
 
   <PropertyGroup>
     <OutputType>Exe</OutputType>
-    <TargetFramework>net8.0</TargetFramework>
+    <TargetFramework>net10.0</TargetFramework>
     <ImplicitUsings>enable</ImplicitUsings>
     <Nullable>enable</Nullable>
     <Configuration>Release</Configuration>
diff --git a/LLama.Examples/Examples/MtmdInteractiveModeExecute.cs b/LLama.Examples/Examples/MtmdInteractiveModeExecute.cs
index 571a89c1a..b93e532f8 100644
--- a/LLama.Examples/Examples/MtmdInteractiveModeExecute.cs
+++ b/LLama.Examples/Examples/MtmdInteractiveModeExecute.cs
@@ -1,11 +1,7 @@
-using System.Collections.Generic;
-using System.IO;
 using System.Text;
 using System.Text.RegularExpressions;
-using LLama;
 using LLama.Common;
 using LLama.Exceptions;
-using Spectre.Console;
 using LLama.Native;
 using LLama.Sampling;
 
@@ -17,11 +13,11 @@ public class MtmdInteractiveModeExecute
     {
         public static async Task Run()
         {
-            string multiModalProj = UserSettings.GetMMProjPath();
-            string modelPath = UserSettings.GetModelPath();
+            var multiModalProj = UserSettings.GetMMProjPath();
+            var modelPath = UserSettings.GetModelPath();
             const int maxTokens = 4096;
 
-            string? prompt = await File.ReadAllTextAsync("Assets/chat-with-bob.json");
+            var prompt = await File.ReadAllTextAsync("Assets/chat-with-bob.json");
 
             var parameters = new ModelParams(modelPath);
 
@@ -164,7 +160,7 @@ void ResetConversation()
                             if (isImage)
                             {
                                 // Keep the raw image data so the caller can reuse or inspect the images later.
-                                imageList.Add(File.ReadAllBytes(mediaPath));
+                                imageList.Add(await File.ReadAllBytesAsync(mediaPath));
                             }
                             else if (isAudio)
                             {
@@ -197,18 +193,7 @@ void ResetConversation()
                     if (imageList.Count > 0)
                     {
                         Console.ForegroundColor = ConsoleColor.Yellow;
-                        Console.WriteLine("Here are the images, that are sent to the chat model in addition to your message.");
-                        Console.WriteLine();
-
-                        foreach (var consoleImage in imageList.Select(image => new CanvasImage(image.ToArray())))
-                        {
-                            consoleImage.MaxWidth = 50;
-                            AnsiConsole.Write(consoleImage);
-                        }
-
-                        Console.WriteLine();
-                        Console.ForegroundColor = ConsoleColor.Yellow;
-                        Console.WriteLine("The images were scaled down for the console only, the model gets full versions.");
+                        Console.WriteLine($"Loaded images files: {string.Join(", ", imageList)}");
                         Console.WriteLine();
                     }
 
diff --git a/LLama.Examples/LLama.Examples.csproj b/LLama.Examples/LLama.Examples.csproj
index bc3f3abec..32f55cb63 100644
--- a/LLama.Examples/LLama.Examples.csproj
+++ b/LLama.Examples/LLama.Examples.csproj
@@ -2,7 +2,7 @@
   <Import Project="..\LLama\LLamaSharp.Runtime.targets" />
   <PropertyGroup>
     <OutputType>Exe</OutputType>
-    <TargetFrameworks>net8.0</TargetFrameworks>
+    <TargetFrameworks>net10.0</TargetFrameworks>
     <ImplicitUsings>enable</ImplicitUsings>
     <Nullable>enable</Nullable>
     <Platforms>AnyCPU;x64</Platforms>
@@ -14,15 +14,13 @@
   </PropertyGroup>
 
   <ItemGroup>
-    <PackageReference Include="Microsoft.Extensions.Logging.Console" Version="10.0.5" />
-    <PackageReference Include="NAudio" Version="2.3.0" />
-    <PackageReference Include="SixLabors.ImageSharp" Version="3.1.12" />
-    <PackageReference Include="Spectre.Console" Version="0.54.0" />
-    <PackageReference Include="Spectre.Console.ImageSharp" Version="0.54.0" />
-    <PackageReference Include="Whisper.net" Version="1.9.0" />
-    <PackageReference Include="Whisper.net.Runtime" Version="1.9.0" />
+    <PackageReference Include="Microsoft.Extensions.Logging.Console" Version="10.0.13" />
+    <PackageReference Include="NAudio" Version="3.1.0" />
+    <PackageReference Include="Spectre.Console" Version="0.57.2" />
+    <PackageReference Include="Whisper.net" Version="1.9.1" />
+    <PackageReference Include="Whisper.net.Runtime" Version="1.9.1" />
     <PackageReference Include="Whisper.net.Runtime.Clblast" Version="1.5.0" />
-    <PackageReference Include="Whisper.net.Runtime.CoreML" Version="1.9.0" />
+    <PackageReference Include="Whisper.net.Runtime.CoreML" Version="1.9.1" />
     <PackageReference Include="Whisper.net.Runtime.Cublas" Version="1.5.0" />
     <PackageReference Include="Microsoft.Agents.AI" Version="*-*" />
     <PackageReference Include="Microsoft.Extensions.AI" Version="*-*" />
diff --git a/LLama.Unittest/LLama.Unittest.csproj b/LLama.Unittest/LLama.Unittest.csproj
index ae7c6d934..447d41387 100644
--- a/LLama.Unittest/LLama.Unittest.csproj
+++ b/LLama.Unittest/LLama.Unittest.csproj
@@ -1,7 +1,7 @@
 <Project Sdk="Microsoft.NET.Sdk">
   <Import Project="..\LLama\LLamaSharp.Runtime.targets" />
   <PropertyGroup>
-    <TargetFramework>net8.0</TargetFramework>
+    <TargetFramework>net10.0</TargetFramework>
     <RootNamespace>LLama.Unittest</RootNamespace>
     <ImplicitUsings>enable</ImplicitUsings>
     <Platforms>AnyCPU;x64</Platforms>
@@ -13,10 +13,10 @@
   </PropertyGroup>
 
   <ItemGroup>
-    <PackageReference Include="Microsoft.NET.Test.Sdk" Version="18.3.0" />
-    <PackageReference Include="Moq" Version="4.20.72" />
+    <PackageReference Include="Microsoft.NET.Test.Sdk" Version="18.11.0" />
+    <PackageReference Include="Moq" Version="4.21.0" />
     <PackageReference Include="xunit" Version="2.9.3" />
-    <PackageReference Include="xunit.runner.visualstudio" Version="3.1.5">
+    <PackageReference Include="xunit.runner.visualstudio" Version="4.0.1">
       <IncludeAssets>runtime; build; native; contentfiles; analyzers; buildtransitive</IncludeAssets>
       <PrivateAssets>all</PrivateAssets>
     </PackageReference>
@@ -24,7 +24,7 @@
       <IncludeAssets>runtime; build; native; contentfiles; analyzers; buildtransitive</IncludeAssets>
       <PrivateAssets>all</PrivateAssets>
     </PackageReference>
-    <PackageReference Include="Xunit.SkippableFact" Version="1.5.61" />
+    <PackageReference Include="Xunit.SkippableFact" Version="1.5.85" />
   </ItemGroup>
 
     <!-- Define each file to download.
diff --git a/LLama.Web/LLama.Web.csproj b/LLama.Web/LLama.Web.csproj
index e09447c61..a8251e66e 100644
--- a/LLama.Web/LLama.Web.csproj
+++ b/LLama.Web/LLama.Web.csproj
@@ -1,7 +1,7 @@
 <Project Sdk="Microsoft.NET.Sdk.Web">
   <Import Project="..\LLama\LLamaSharp.Runtime.targets" />
   <PropertyGroup>
-    <TargetFramework>net8.0</TargetFramework>
+    <TargetFramework>net10.0</TargetFramework>
     <Nullable>disable</Nullable>
     <ImplicitUsings>enable</ImplicitUsings>
   </PropertyGroup>
@@ -15,7 +15,7 @@
   </ItemGroup>
 
   <ItemGroup>
-    <PackageReference Include="Microsoft.AspNetCore.Mvc.Razor.RuntimeCompilation" Version="8.0.31" />
+    <PackageReference Include="Microsoft.AspNetCore.Mvc.Razor.RuntimeCompilation" Version="10.0.13" />
   </ItemGroup>
 
 </Project>
diff --git a/LLama.WebAPI/LLama.WebAPI.csproj b/LLama.WebAPI/LLama.WebAPI.csproj
index 43fa14d0f..a5b71599b 100644
--- a/LLama.WebAPI/LLama.WebAPI.csproj
+++ b/LLama.WebAPI/LLama.WebAPI.csproj
@@ -1,7 +1,7 @@
 <Project Sdk="Microsoft.NET.Sdk.Web">
   <Import Project="..\LLama\LLamaSharp.Runtime.targets" />
   <PropertyGroup>
-    <TargetFramework>net8.0</TargetFramework>
+    <TargetFramework>net10.0</TargetFramework>
     <Nullable>enable</Nullable>
     <ImplicitUsings>enable</ImplicitUsings>
     <InvariantGlobalization>true</InvariantGlobalization>
diff --git a/LLama/LLamaSharp.csproj b/LLama/LLamaSharp.csproj
index 7a5e3db2e..9acd98bfa 100644
--- a/LLama/LLamaSharp.csproj
+++ b/LLama/LLamaSharp.csproj
@@ -1,6 +1,6 @@
 <Project Sdk="Microsoft.NET.Sdk">
   <PropertyGroup>
-    <TargetFrameworks>netstandard2.0;net8.0</TargetFrameworks>
+    <TargetFrameworks>netstandard2.0;net10.0</TargetFrameworks>
     <RootNamespace>LLama</RootNamespace>
     <Nullable>enable</Nullable>
     <LangVersion>13</LangVersion>
@@ -44,18 +44,18 @@
   <ItemGroup Condition="'$(TargetFramework)' == 'netstandard2.0'">
     <PackageReference Include="IsExternalInit" Version="1.0.3" PrivateAssets="all" />
     <PackageReference Include="System.Memory" Version="4.6.3" PrivateAssets="all" />
-    <PackageReference Include="System.Text.Json" Version="10.0.10" />
+    <PackageReference Include="System.Text.Json" Version="10.0.13" />
   </ItemGroup>
 
   <ItemGroup>
     <PackageReference Include="CommunityToolkit.HighPerformance" Version="8.4.2" />
-    <PackageReference Include="Microsoft.Bcl.AsyncInterfaces" Version="10.0.10" />
-    <PackageReference Include="Microsoft.Bcl.Memory" Version="10.0.12" />
-    <PackageReference Include="Microsoft.Extensions.AI.Abstractions" Version="10.8.3" />
-    <PackageReference Include="Microsoft.Extensions.Logging.Abstractions" Version="10.0.10" />
-    <PackageReference Include="System.Linq.Async" Version="7.0.0" />
-    <PackageReference Include="System.Linq.AsyncEnumerable" Version="10.0.2" />
-    <PackageReference Include="System.Numerics.Tensors" Version="10.0.5" />
+    <PackageReference Include="Microsoft.Bcl.AsyncInterfaces" Version="10.0.13" />
+    <PackageReference Include="Microsoft.Bcl.Memory" Version="10.0.13" />
+    <PackageReference Include="Microsoft.Extensions.AI.Abstractions" Version="10.11.0" />
+    <PackageReference Include="Microsoft.Extensions.Logging.Abstractions" Version="10.0.13" />
+    <PackageReference Include="System.Linq.Async" Version="7.0.1" />
+    <PackageReference Include="System.Linq.AsyncEnumerable" Version="10.0.13" />
+    <PackageReference Include="System.Numerics.Tensors" Version="10.0.13" />
   </ItemGroup>
 
   <PropertyGroup>

From e3775b6dbede7a34aa298c40557f32b7aac10759 Mon Sep 17 00:00:00 2001
From: Martin Evans <martindevans@gmail.com>
Date: Wed, 14 Oct 2026 23:54:44 +0100
Subject: [PATCH 2/5] - Commented out SpeechChat.cs example, currently broken
 by naudio upgrade  - Added lots of uses of cancellation token throughout test
 code to resolve warnings  - Added some comments to remove more warnings

---
 LLama.Examples/ExampleRunner.cs               |   2 +-
 .../Examples/BatchedExecutorMtmd.cs           |   1 -
 LLama.Examples/Examples/SpeechChat.cs         | 475 +++++++++---------
 LLama.Unittest/AssemblyAttributes.cs          |   6 +-
 LLama.Unittest/BasicTest.cs                   |   1 -
 LLama.Unittest/LLama.Unittest.csproj          |   7 +-
 LLama.Unittest/LLamaEmbedderTests.cs          |   1 -
 .../LLamaExecutorChatClientTests.cs           |  68 ++-
 LLama.Unittest/LLamaRerankerTests.cs          |   8 +-
 LLama.Unittest/MtmdExecutorTests.cs           |  13 +-
 LLama.Unittest/SamplingTests.cs               |   2 -
 LLama.Unittest/StatelessExecutorTest.cs       |  13 +-
 LLama.Unittest/StreamingTextDecoderTests.cs   |   1 -
 LLama.Unittest/TemplateTests.cs               |   3 -
 LLama.Unittest/TextTransformTests.cs          |   7 +-
 LLama.Unittest/TokenTests.cs                  |  10 +
 LLama/Extensions/IModelParamsExtensions.cs    |   2 +-
 LLama/Extensions/IntPtrExtensions.cs          |   4 +-
 LLama/LLamaInstructExecutor.cs                |  10 +-
 LLama/LLamaInteractExecutor.cs                |   8 +-
 LLama/Native/NativeApi.cs                     |   8 +
 LLama/Native/SafeMtmdModelHandle.cs           |   2 +-
 22 files changed, 341 insertions(+), 311 deletions(-)

diff --git a/LLama.Examples/ExampleRunner.cs b/LLama.Examples/ExampleRunner.cs
index e3ffdf67c..53946a2e6 100644
--- a/LLama.Examples/ExampleRunner.cs
+++ b/LLama.Examples/ExampleRunner.cs
@@ -32,7 +32,7 @@ public class ExampleRunner
         { "Batched Executor: BoolQ Benchmark", BatchedExecutorBoolQ.Run },
         { "Batched Executor: Beam Search", BatchedExecutorBeamSearch.Run },
         { "Custom Sampling Pipeline", CustomSampler.Run },
-        { "Speech Chat: Integration with Whisper.net", SpeechChat.Run },
+        //{ "Speech Chat: Integration with Whisper.net", SpeechChat.Run },
         { "Exit", () => { Environment.Exit(0); return Task.CompletedTask; } }
     };
 
diff --git a/LLama.Examples/Examples/BatchedExecutorMtmd.cs b/LLama.Examples/Examples/BatchedExecutorMtmd.cs
index 1d29c170f..29b32c2aa 100644
--- a/LLama.Examples/Examples/BatchedExecutorMtmd.cs
+++ b/LLama.Examples/Examples/BatchedExecutorMtmd.cs
@@ -40,7 +40,6 @@ public static async Task Run()
         var promptText = string.Concat(marker, promptSuffix);
 
         var imagePath = UserSettings.GetImagePath();
-        AnsiConsole.Write(new CanvasImage(imagePath));
 
         var vocab = executor.Context.NativeHandle.ModelHandle.Vocab;
 
diff --git a/LLama.Examples/Examples/SpeechChat.cs b/LLama.Examples/Examples/SpeechChat.cs
index e12aabe94..596970eae 100644
--- a/LLama.Examples/Examples/SpeechChat.cs
+++ b/LLama.Examples/Examples/SpeechChat.cs
@@ -1,239 +1,240 @@
-using LLama.Common;
-using NAudio.Wave;
-using Spectre.Console;
-using Whisper.net;
-
-namespace LLama.Examples.Examples
-{
-    public static class SpeechChat
-    {
-        public static async Task Run()
-        {
-            AnsiConsole.MarkupLine("""
-                                  [yellow on black]
-                                  This example demonstrates the basics of audio transcriptions, speech recognition, and speech commands,
-                                      as well as how to recognize a user's voice in real time and then get a response from LLM.
-                                  It uses whisper.net and models could be found in: https://huggingface.co/ggerganov/whisper.cpp/tree/main.
-                                  To use it, you need a working microphone and enough RAM to host both audio + language models.
-                                  Once you've selected the models, just speak to your microphone and watch the LLM continue your text.
-                                  While it's going, you can say something like 'Okay, stop', or 'Stop now', to interrupt the LLM's inference.
+//using LLama.Common;
+//using NAudio.Wave;
+//using Spectre.Console;
+//using Whisper.net;
+
+//namespace LLama.Examples.Examples
+//{
+//    public static class SpeechChat
+//    {
+//        public static async Task Run()
+//        {
+//            AnsiConsole.MarkupLine("""
+//                                  [yellow on black]
+//                                  This example demonstrates the basics of audio transcriptions, speech recognition, and speech commands,
+//                                      as well as how to recognize a user's voice in real time and then get a response from LLM.
+//                                  It uses whisper.net and models could be found in: https://huggingface.co/ggerganov/whisper.cpp/tree/main.
+//                                  To use it, you need a working microphone and enough RAM to host both audio + language models.
+//                                  Once you've selected the models, just speak to your microphone and watch the LLM continue your text.
+//                                  While it's going, you can say something like 'Okay, stop', or 'Stop now', to interrupt the LLM's inference.
                                   
-                                  NOTE: You may need to poke around with the voice detection threshold, based on your mic's sensitivity.
-                                  [/]
-                                  """);
-
-            AnsiConsole.MarkupLine("[white on black]You can find the official ggml models in whisper.cpp's huggingface repository: https://huggingface.co/ggerganov/whisper.cpp/tree/main [/]");
-
-            var whisperModel = UserSettings.GetWhisperPath();
-            var languageModel = UserSettings.GetModelPath();
-
-            using var speechRecognitionServer = new SpeechRecognitionServer(whisperModel);
-
-            using var _ = new LlamaSessionSpeechListener(speechRecognitionServer, languageModel);
-
-            AnsiConsole.MarkupLine("[green]Voice active. Begin talking to transcribe. Press any key at any time to exit.[/]");
-            await Task.Delay(1000);
-            Console.ReadKey();
-        }
-
-
-        private class LlamaSessionSpeechListener
-            : ISpeechListener, IDisposable
-        {
-            private bool _isModelResponding;
-            private readonly SpeechRecognitionServer _audioServer;
-
-            private readonly LLamaWeights _model;
-            private readonly LLamaContext _context;
-            private readonly InteractiveExecutor _executor;
-
-            private string _fullPrompt = "";
-            private bool _canceled;
-
-            public LlamaSessionSpeechListener(SpeechRecognitionServer server, string languageModelPath)
-            {
-                var parameters = new ModelParams(languageModelPath)
-                {
-                    GpuLayerCount = 99
-                };
-                _model = LLamaWeights.LoadFromFile(parameters);
-                _context = _model.CreateContext(parameters);
-                _executor = new InteractiveExecutor(_context);
-
-                _audioServer = server;
-                _audioServer.ServiceUsers.Add(this);
-            }
-
-            // Whisper is struggling with single words and very short phrases without context, so it's actually better to say something like "Ok, Stop!" to have it work better.
-            bool ISpeechListener.IsInterested(string audioTranscription) => !_isModelResponding || audioTranscription.Contains("stop", StringComparison.CurrentCultureIgnoreCase);
-            void ISpeechListener.HandleSpeech(string audioTranscription)
-            {
-                if (_isModelResponding && audioTranscription.Contains("stop", StringComparison.CurrentCultureIgnoreCase))
-                    _canceled = true;
-                else if (!_isModelResponding)
-                    _ = SendMessage(audioTranscription);
-            }
-
-            private async Task SendMessage(string newMessage)
-            {
-                // While a response is queried, we want to detect short phrases/commands like 'stop',
-                _audioServer.DetectionSettings = (1, 2); // ..so we lower the min Speech Detection time.
-
-                _isModelResponding = true;
-                AddToPrompt($"\n{newMessage}\n", ConsoleColor.Blue);
-                await foreach (var token in _executor.InferAsync(_fullPrompt))
-                {
-                    AddToPrompt(token);
-                    if (_canceled)
-                    {
-                        AddToPrompt("[...stopped]", ConsoleColor.Red);
-                        break;
-                    }
-                }
-                _audioServer.DetectionSettings = (2, 3);         // Reset back to default detection settings to avoid false positives.
-                (_isModelResponding, _canceled) = (false, false); // Reset the state variables to their default.
-            }
-
-            private void AddToPrompt(string msg, ConsoleColor color = ConsoleColor.Yellow)
-            {
-                _fullPrompt += msg;
-
-                AnsiConsole.Markup($"[{color}]{Markup.Escape(msg)}[/]");
-            }
-
-            void IDisposable.Dispose()
-            {
-                _model.Dispose();
-                _context.Dispose();
-            }
-        }
-
-        public interface ISpeechListener
-        {
-            bool IsInterested(string audioTranscription);
-            void HandleSpeech(string audioTranscription);
-        }
-
-        public sealed class SpeechRecognitionServer
-            : IDisposable
-        {
-            private static readonly TimeSpan ClipLength = TimeSpan.FromMilliseconds(250);
-
-            private const float VoiceDetectionThreshold = 0.01f; // Adjust as needed
-            private readonly string[] _knownFalsePositives = ["[BLANK_AUDIO]", "Thank you", "[silence]"];
-
-            private readonly WaveInEvent _waveIn;
-            private readonly WaveFormat _waveFormat = new(16000, 16, 1); // 16KHz, 16 bits, Mono Channel
-            private readonly List<byte> _recordedBytes = [];
-
-            private readonly WhisperProcessor? _processor;
-
-            private readonly string _whisperPrompt =
-"""
-The short audio comes from a user that is speaking to an AI Language Model in real time.
-Pay extra attentions for commands like 'ok stop' or just 'stop'.
-In case of inaudible sentences that might be, assume they're saying 'stop'.
-""".Trim();
-
-            // Tracked stats for Speech Recognition, Parsing, and Serving.
-            private int _currentBlankClips;  // Ideally would work with milliseconds,
-            private int _totalNonBlankClips; // ..but for example's sake they work on a
-
-            private int _nonIdleTime;        // ..clip-based quant-length (1 = clipLength).
-            // Default detection settings: A speech of 750ms, followed by pause of 500ms. (2x250ms)
-            public (int minBlanksPerSeparation, int minNonBlanksForValidMessages) DetectionSettings = (2, 3);
-
-            public readonly HashSet<ISpeechListener> ServiceUsers = [];
-
-            public SpeechRecognitionServer(string modelPath)
-            {
-                var whisperFactory = WhisperFactory.FromPath(modelPath);
-
-                var builder = whisperFactory
-                    .CreateBuilder()
-                    .WithThreads(16)
-                    .WithPrompt(_whisperPrompt)
-                    .WithSingleSegment()
-                    .WithLanguage("en");
-                ((BeamSearchSamplingStrategyBuilder)builder.WithBeamSearchSamplingStrategy()).WithPatience(0.2f).WithBeamSize(5);
-                _processor = builder.Build();
-
-                _waveIn = new WaveInEvent { BufferMilliseconds = (int)ClipLength.TotalMilliseconds, WaveFormat = _waveFormat };
-                _waveIn.DataAvailable += OnAudioDataAvailable;
-                _waveIn.StartRecording();
-            }
-
-            private void OnAudioDataAvailable(object? sender, WaveInEventArgs e)
-            {
-                // Cache the recorded bytes
-                _recordedBytes.AddRange(e.Buffer[..e.BytesRecorded]);
-                if (_recordedBytes.Count > 110000000) { _recordedBytes.RemoveRange(0, 50000000); }
-
-                // Get the max volume contained inside the clip. Since the clip is recorded as bytes, we need to translate them to samples before getting their volume.
-                var maxVolume = 0f; // This byte->sample algorithm is from: https://github.com/naudio/NAudio/blob/master/Docs/RecordingLevelMeter.md#calculating-peak-values
-                for (var i = 0; i < e.BytesRecorded; i += 2)
-                {
-                    maxVolume = Math.Max(maxVolume, Math.Abs((short)((e.Buffer[i + 1] << 8) | e.Buffer[i + 0]) / 32768f));
-                }
-
-                // Compare the volume with the threshold and act accordingly. Once an interesting and 'full' set of clips pops up, serve it.
-                if (maxVolume >= VoiceDetectionThreshold)
-                {
-                    _currentBlankClips = 0;
-                    _totalNonBlankClips++;
-                    _nonIdleTime++;
-                }
-                else if (++_currentBlankClips < DetectionSettings.minBlanksPerSeparation)
-                {
-                    _nonIdleTime++;
-                }
-                else
-                {
-                    if (_totalNonBlankClips >= DetectionSettings.minNonBlanksForValidMessages)
-                        SendTranscription();
-                    else if (_totalNonBlankClips > 0) { } // This might be case of a false-positive -- knock, noise, cough, anything.
-                    (_currentBlankClips, _totalNonBlankClips, _nonIdleTime) = (0, 0, 0);
-                }
-
-
-                async void SendTranscription()
-                {
-                    var bytesPerClip = _waveFormat.BitsPerSample * (int)ClipLength.TotalMilliseconds * 2;
-                    var capturedClipBytes = _recordedBytes.TakeLast(bytesPerClip * (_nonIdleTime + 2)).ToArray();
-
-                    // Save to temporary file.
-                    var transcribedText = await ProcessAudio(capturedClipBytes, "Assets\\temp.wav"); 
-
-                    // False positive.. yikes!
-                    if (_knownFalsePositives.Contains(transcribedText))
-                        return;    
+//                                  NOTE: You may need to poke around with the voice detection threshold, based on your mic's sensitivity.
+//                                  [/]
+//                                  """);
+
+//            AnsiConsole.MarkupLine("[white on black]You can find the official ggml models in whisper.cpp's huggingface repository: https://huggingface.co/ggerganov/whisper.cpp/tree/main [/]");
+
+//            var whisperModel = UserSettings.GetWhisperPath();
+//            var languageModel = UserSettings.GetModelPath();
+
+//            using var speechRecognitionServer = new SpeechRecognitionServer(whisperModel);
+
+//            using var _ = new LlamaSessionSpeechListener(speechRecognitionServer, languageModel);
+
+//            AnsiConsole.MarkupLine("[green]Voice active. Begin talking to transcribe. Press any key at any time to exit.[/]");
+//            await Task.Delay(1000);
+//            Console.ReadKey();
+//        }
+
+
+//        private class LlamaSessionSpeechListener
+//            : ISpeechListener, IDisposable
+//        {
+//            private bool _isModelResponding;
+//            private readonly SpeechRecognitionServer _audioServer;
+
+//            private readonly LLamaWeights _model;
+//            private readonly LLamaContext _context;
+//            private readonly InteractiveExecutor _executor;
+
+//            private string _fullPrompt = "";
+//            private bool _canceled;
+
+//            public LlamaSessionSpeechListener(SpeechRecognitionServer server, string languageModelPath)
+//            {
+//                var parameters = new ModelParams(languageModelPath)
+//                {
+//                    GpuLayerCount = 99
+//                };
+//                _model = LLamaWeights.LoadFromFile(parameters);
+//                _context = _model.CreateContext(parameters);
+//                _executor = new InteractiveExecutor(_context);
+
+//                _audioServer = server;
+//                _audioServer.ServiceUsers.Add(this);
+//            }
+
+//            // Whisper is struggling with single words and very short phrases without context, so it's actually better to say something like "Ok, Stop!" to have it work better.
+//            bool ISpeechListener.IsInterested(string audioTranscription) => !_isModelResponding || audioTranscription.Contains("stop", StringComparison.CurrentCultureIgnoreCase);
+//            void ISpeechListener.HandleSpeech(string audioTranscription)
+//            {
+//                if (_isModelResponding && audioTranscription.Contains("stop", StringComparison.CurrentCultureIgnoreCase))
+//                    _canceled = true;
+//                else if (!_isModelResponding)
+//                    _ = SendMessage(audioTranscription);
+//            }
+
+//            private async Task SendMessage(string newMessage)
+//            {
+//                // While a response is queried, we want to detect short phrases/commands like 'stop',
+//                _audioServer.DetectionSettings = (1, 2); // ..so we lower the min Speech Detection time.
+
+//                _isModelResponding = true;
+//                AddToPrompt($"\n{newMessage}\n", ConsoleColor.Blue);
+//                await foreach (var token in _executor.InferAsync(_fullPrompt))
+//                {
+//                    AddToPrompt(token);
+//                    if (_canceled)
+//                    {
+//                        AddToPrompt("[...stopped]", ConsoleColor.Red);
+//                        break;
+//                    }
+//                }
+//                _audioServer.DetectionSettings = (2, 3);         // Reset back to default detection settings to avoid false positives.
+//                (_isModelResponding, _canceled) = (false, false); // Reset the state variables to their default.
+//            }
+
+//            private void AddToPrompt(string msg, ConsoleColor color = ConsoleColor.Yellow)
+//            {
+//                _fullPrompt += msg;
+
+//                AnsiConsole.Markup($"[{color}]{Markup.Escape(msg)}[/]");
+//            }
+
+//            void IDisposable.Dispose()
+//            {
+//                _model.Dispose();
+//                _context.Dispose();
+//            }
+//        }
+
+//        public interface ISpeechListener
+//        {
+//            bool IsInterested(string audioTranscription);
+//            void HandleSpeech(string audioTranscription);
+//        }
+
+//        public sealed class SpeechRecognitionServer
+//            : IDisposable
+//        {
+//            private static readonly TimeSpan ClipLength = TimeSpan.FromMilliseconds(250);
+
+//            private const float VoiceDetectionThreshold = 0.01f; // Adjust as needed
+//            private readonly string[] _knownFalsePositives = ["[BLANK_AUDIO]", "Thank you", "[silence]"];
+
+//            private readonly WaveInEvent _waveIn;
+
+//            private readonly WaveFormat _waveFormat = new(16000, 16, 1); // 16KHz, 16 bits, Mono Channel
+//            private readonly List<byte> _recordedBytes = [];
+
+//            private readonly WhisperProcessor? _processor;
+
+//            private readonly string _whisperPrompt =
+//"""
+//The short audio comes from a user that is speaking to an AI Language Model in real time.
+//Pay extra attentions for commands like 'ok stop' or just 'stop'.
+//In case of inaudible sentences that might be, assume they're saying 'stop'.
+//""".Trim();
+
+//            // Tracked stats for Speech Recognition, Parsing, and Serving.
+//            private int _currentBlankClips;  // Ideally would work with milliseconds,
+//            private int _totalNonBlankClips; // ..but for example's sake they work on a
+
+//            private int _nonIdleTime;        // ..clip-based quant-length (1 = clipLength).
+//            // Default detection settings: A speech of 750ms, followed by pause of 500ms. (2x250ms)
+//            public (int minBlanksPerSeparation, int minNonBlanksForValidMessages) DetectionSettings = (2, 3);
+
+//            public readonly HashSet<ISpeechListener> ServiceUsers = [];
+
+//            public SpeechRecognitionServer(string modelPath)
+//            {
+//                var whisperFactory = WhisperFactory.FromPath(modelPath);
+
+//                var builder = whisperFactory
+//                    .CreateBuilder()
+//                    .WithThreads(16)
+//                    .WithPrompt(_whisperPrompt)
+//                    .WithSingleSegment()
+//                    .WithLanguage("en");
+//                ((BeamSearchSamplingStrategyBuilder)builder.WithBeamSearchSamplingStrategy()).WithPatience(0.2f).WithBeamSize(5);
+//                _processor = builder.Build();
+
+//                _waveIn = new WaveInEvent { BufferMilliseconds = (int)ClipLength.TotalMilliseconds, WaveFormat = _waveFormat };
+//                _waveIn.DataAvailable += OnAudioDataAvailable;
+//                _waveIn.StartRecording();
+//            }
+
+//            private void OnAudioDataAvailable(object? sender, WaveInEventArgs e)
+//            {
+//                // Cache the recorded bytes
+//                _recordedBytes.AddRange(e.Buffer[..e.BytesRecorded]);
+//                if (_recordedBytes.Count > 110000000) { _recordedBytes.RemoveRange(0, 50000000); }
+
+//                // Get the max volume contained inside the clip. Since the clip is recorded as bytes, we need to translate them to samples before getting their volume.
+//                var maxVolume = 0f; // This byte->sample algorithm is from: https://github.com/naudio/NAudio/blob/master/Docs/RecordingLevelMeter.md#calculating-peak-values
+//                for (var i = 0; i < e.BytesRecorded; i += 2)
+//                {
+//                    maxVolume = Math.Max(maxVolume, Math.Abs((short)((e.Buffer[i + 1] << 8) | e.Buffer[i + 0]) / 32768f));
+//                }
+
+//                // Compare the volume with the threshold and act accordingly. Once an interesting and 'full' set of clips pops up, serve it.
+//                if (maxVolume >= VoiceDetectionThreshold)
+//                {
+//                    _currentBlankClips = 0;
+//                    _totalNonBlankClips++;
+//                    _nonIdleTime++;
+//                }
+//                else if (++_currentBlankClips < DetectionSettings.minBlanksPerSeparation)
+//                {
+//                    _nonIdleTime++;
+//                }
+//                else
+//                {
+//                    if (_totalNonBlankClips >= DetectionSettings.minNonBlanksForValidMessages)
+//                        SendTranscription();
+//                    else if (_totalNonBlankClips > 0) { } // This might be case of a false-positive -- knock, noise, cough, anything.
+//                    (_currentBlankClips, _totalNonBlankClips, _nonIdleTime) = (0, 0, 0);
+//                }
+
+
+//                async void SendTranscription()
+//                {
+//                    var bytesPerClip = _waveFormat.BitsPerSample * (int)ClipLength.TotalMilliseconds * 2;
+//                    var capturedClipBytes = _recordedBytes.TakeLast(bytesPerClip * (_nonIdleTime + 2)).ToArray();
+
+//                    // Save to temporary file.
+//                    var transcribedText = await ProcessAudio(capturedClipBytes, "Assets\\temp.wav"); 
+
+//                    // False positive.. yikes!
+//                    if (_knownFalsePositives.Contains(transcribedText))
+//                        return;    
                     
-                    foreach (var user in ServiceUsers.Where(x => x.IsInterested(transcribedText)))
-                        user.HandleSpeech(transcribedText);
-                }
-            }
-
-            /// <summary> Requests a transcription and responds with the text. </summary>
-            private async Task<string> ProcessAudio(byte[] bytes, string tempWavFilePath)
-            {
-                await using var wavStream = new MemoryStream();
-
-                await using (var writer = new WaveFileWriter(tempWavFilePath, _waveFormat))
-                    writer.Write(bytes, 0, bytes.Length);
-                await using (var fileStream = File.OpenRead(tempWavFilePath))
-                    await fileStream.CopyToAsync(wavStream);
-
-                wavStream.Seek(0, SeekOrigin.Begin);
-
-                Console.Beep();
-                return string.Join(' ', await _processor!.ProcessAsync(wavStream).Select(x => x.Text).ToListAsync()).Trim();
-            }
-
-            void IDisposable.Dispose()
-            {
-                _waveIn.Dispose();
-                _processor?.Dispose();
-            }
-        }
-    }
-}
+//                    foreach (var user in ServiceUsers.Where(x => x.IsInterested(transcribedText)))
+//                        user.HandleSpeech(transcribedText);
+//                }
+//            }
+
+//            /// <summary> Requests a transcription and responds with the text. </summary>
+//            private async Task<string> ProcessAudio(byte[] bytes, string tempWavFilePath)
+//            {
+//                await using var wavStream = new MemoryStream();
+
+//                await using (var writer = new WaveFileWriter(tempWavFilePath, _waveFormat))
+//                    writer.Write(bytes, 0, bytes.Length);
+//                await using (var fileStream = File.OpenRead(tempWavFilePath))
+//                    await fileStream.CopyToAsync(wavStream);
+
+//                wavStream.Seek(0, SeekOrigin.Begin);
+
+//                Console.Beep();
+//                return string.Join(' ', await _processor!.ProcessAsync(wavStream).Select(x => x.Text).ToListAsync()).Trim();
+//            }
+
+//            void IDisposable.Dispose()
+//            {
+//                _waveIn.Dispose();
+//                _processor?.Dispose();
+//            }
+//        }
+//    }
+//}
diff --git a/LLama.Unittest/AssemblyAttributes.cs b/LLama.Unittest/AssemblyAttributes.cs
index 346e4a082..3ba846bc7 100644
--- a/LLama.Unittest/AssemblyAttributes.cs
+++ b/LLama.Unittest/AssemblyAttributes.cs
@@ -1,5 +1,7 @@
-
+using Xunit.Sdk;
+using Xunit.v3;
+
 // Disable running tests in parallel. Some tests load language models and try to and
 // this can cause a lockup due to memory thrashing and terrible performance.
 // e.g. 7GB model loaded by 10 different tests and trying to evaluate prompts
-[assembly: CollectionBehavior(DisableTestParallelization = true)]
+[assembly: Parallelization(Algorithm = ParallelAlgorithm.Conservative, MaxThreads = 1, Mode = ParallelMode.None)]
diff --git a/LLama.Unittest/BasicTest.cs b/LLama.Unittest/BasicTest.cs
index 6624f417f..4bd4602cd 100644
--- a/LLama.Unittest/BasicTest.cs
+++ b/LLama.Unittest/BasicTest.cs
@@ -1,5 +1,4 @@
 using LLama.Common;
-using Xunit.Abstractions;
 
 namespace LLama.Unittest
 {
diff --git a/LLama.Unittest/LLama.Unittest.csproj b/LLama.Unittest/LLama.Unittest.csproj
index 447d41387..e37a615d5 100644
--- a/LLama.Unittest/LLama.Unittest.csproj
+++ b/LLama.Unittest/LLama.Unittest.csproj
@@ -15,16 +15,11 @@
   <ItemGroup>
     <PackageReference Include="Microsoft.NET.Test.Sdk" Version="18.11.0" />
     <PackageReference Include="Moq" Version="4.21.0" />
-    <PackageReference Include="xunit" Version="2.9.3" />
-    <PackageReference Include="xunit.runner.visualstudio" Version="4.0.1">
-      <IncludeAssets>runtime; build; native; contentfiles; analyzers; buildtransitive</IncludeAssets>
-      <PrivateAssets>all</PrivateAssets>
-    </PackageReference>
     <PackageReference Include="coverlet.collector" Version="10.1.0">
       <IncludeAssets>runtime; build; native; contentfiles; analyzers; buildtransitive</IncludeAssets>
       <PrivateAssets>all</PrivateAssets>
     </PackageReference>
-    <PackageReference Include="Xunit.SkippableFact" Version="1.5.85" />
+    <PackageReference Include="xunit.v3" Version="4.0.2" />
   </ItemGroup>
 
     <!-- Define each file to download.
diff --git a/LLama.Unittest/LLamaEmbedderTests.cs b/LLama.Unittest/LLamaEmbedderTests.cs
index 7d7654126..f18a1c68e 100644
--- a/LLama.Unittest/LLamaEmbedderTests.cs
+++ b/LLama.Unittest/LLamaEmbedderTests.cs
@@ -2,7 +2,6 @@
 using LLama.Extensions;
 using LLama.Native;
 using Microsoft.Extensions.AI;
-using Xunit.Abstractions;
 
 namespace LLama.Unittest;
 
diff --git a/LLama.Unittest/LLamaExecutorChatClientTests.cs b/LLama.Unittest/LLamaExecutorChatClientTests.cs
index 95fba1940..99d62a5ee 100644
--- a/LLama.Unittest/LLamaExecutorChatClientTests.cs
+++ b/LLama.Unittest/LLamaExecutorChatClientTests.cs
@@ -14,14 +14,18 @@ public async Task RawRepresentationSettingsAreKept()
     {
         var executor = new CapturingExecutor();
 
-        await executor.AsChatClient().GetResponseAsync("hello", new ChatOptions
-        {
-            RawRepresentationFactory = _ => new InferenceParams
+        await executor.AsChatClient().GetResponseAsync(
+            "hello",
+            new ChatOptions
             {
-                MaxTokens = 1024,
-                SamplingPipeline = new DefaultSamplingPipeline { Temperature = 0.1f, TopP = 0.5f, TopK = 5, Seed = 42 },
+                RawRepresentationFactory = _ => new InferenceParams
+                {
+                    MaxTokens = 1024,
+                    SamplingPipeline = new DefaultSamplingPipeline { Temperature = 0.1f, TopP = 0.5f, TopK = 5, Seed = 42 },
+                }
             },
-        });
+            cancellationToken: TestContext.Current.CancellationToken
+        );
 
         var pipeline = Assert.IsType<DefaultSamplingPipeline>(executor.InferenceParams?.SamplingPipeline);
         Assert.Equal(1024, executor.InferenceParams!.MaxTokens);
@@ -36,10 +40,14 @@ public async Task RawRepresentationUnlimitedMaxTokensIsKept()
     {
         var executor = new CapturingExecutor();
 
-        await executor.AsChatClient().GetResponseAsync("hello", new ChatOptions
-        {
-            RawRepresentationFactory = _ => new InferenceParams { MaxTokens = -1 },
-        });
+        await executor.AsChatClient().GetResponseAsync(
+            "hello",
+            new ChatOptions
+            {
+                RawRepresentationFactory = _ => new InferenceParams { MaxTokens = -1 },
+            },
+            cancellationToken: TestContext.Current.CancellationToken
+        );
 
         Assert.Equal(-1, executor.InferenceParams!.MaxTokens);
     }
@@ -49,19 +57,23 @@ public async Task ChatOptionsOverrideRawRepresentationSettings()
     {
         var executor = new CapturingExecutor();
 
-        await executor.AsChatClient().GetResponseAsync("hello", new ChatOptions
-        {
-            MaxOutputTokens = 64,
-            Temperature = 0.9f,
-            TopP = 0.8f,
-            TopK = 20,
-            Seed = 7,
-            RawRepresentationFactory = _ => new InferenceParams
+        await executor.AsChatClient().GetResponseAsync(
+            "hello",
+            new ChatOptions
             {
-                MaxTokens = 1024,
-                SamplingPipeline = new DefaultSamplingPipeline { Temperature = 0.1f, TopP = 0.5f, TopK = 5, Seed = 42 },
+                MaxOutputTokens = 64,
+                Temperature = 0.9f,
+                TopP = 0.8f,
+                TopK = 20,
+                Seed = 7,
+                RawRepresentationFactory = _ => new InferenceParams
+                {
+                    MaxTokens = 1024,
+                    SamplingPipeline = new DefaultSamplingPipeline { Temperature = 0.1f, TopP = 0.5f, TopK = 5, Seed = 42 },
+                },
             },
-        });
+            cancellationToken: TestContext.Current.CancellationToken
+        );
 
         var pipeline = Assert.IsType<DefaultSamplingPipeline>(executor.InferenceParams?.SamplingPipeline);
         Assert.Equal(64, executor.InferenceParams!.MaxTokens);
@@ -77,10 +89,14 @@ public async Task RawRepresentationSamplingPipelineIsKept()
         var executor = new CapturingExecutor();
         var greedy = new GreedySamplingPipeline();
 
-        await executor.AsChatClient().GetResponseAsync("hello", new ChatOptions
-        {
-            RawRepresentationFactory = _ => new InferenceParams { SamplingPipeline = greedy },
-        });
+        await executor.AsChatClient().GetResponseAsync(
+            "hello",
+            new ChatOptions
+            {
+                RawRepresentationFactory = _ => new InferenceParams { SamplingPipeline = greedy },
+            },
+            cancellationToken: TestContext.Current.CancellationToken
+        );
 
         Assert.Same(greedy, executor.InferenceParams?.SamplingPipeline);
     }
@@ -90,7 +106,7 @@ public async Task MaxTokensDefaultsTo256WithoutALimit()
     {
         var executor = new CapturingExecutor();
 
-        await executor.AsChatClient().GetResponseAsync("hello");
+        await executor.AsChatClient().GetResponseAsync("hello", cancellationToken: TestContext.Current.CancellationToken);
 
         Assert.Equal(256, executor.InferenceParams!.MaxTokens);
     }
diff --git a/LLama.Unittest/LLamaRerankerTests.cs b/LLama.Unittest/LLamaRerankerTests.cs
index 9ba0a31c0..12022dcdc 100644
--- a/LLama.Unittest/LLamaRerankerTests.cs
+++ b/LLama.Unittest/LLamaRerankerTests.cs
@@ -1,9 +1,5 @@
 using LLama.Common;
-using LLama.Extensions;
 using LLama.Native;
-using Microsoft.Extensions.AI;
-using System.Runtime.InteropServices;
-using Xunit.Abstractions;
 
 namespace LLama.Unittest;
 
@@ -43,7 +39,7 @@ public async Task CompareRerankingScore()
             "sometimes called a panda bear or simply panda",
             "is a bear species endemic to China.") 
         };
-        var scores = await _reranker.GetRelevanceScores(input, documents, normalize: false);
+        var scores = await _reranker.GetRelevanceScores(input, documents, normalize: false, cancellationToken: TestContext.Current.CancellationToken);
 
         Assert.True(documents.Length == scores.Count);
 
@@ -63,7 +59,7 @@ public async Task MostRelevantDocument()
             "sometimes called a panda bear or simply panda",
             "is a bear species endemic to China.")
         };
-        var scores = await _reranker.GetRelevanceScores(input, documents, normalize: true);
+        var scores = await _reranker.GetRelevanceScores(input, documents, normalize: true, TestContext.Current.CancellationToken);
 
         Assert.NotNull(scores);
         Assert.True(documents.Length == scores.Count);
diff --git a/LLama.Unittest/MtmdExecutorTests.cs b/LLama.Unittest/MtmdExecutorTests.cs
index 7aa89e2f9..a8e83cc52 100644
--- a/LLama.Unittest/MtmdExecutorTests.cs
+++ b/LLama.Unittest/MtmdExecutorTests.cs
@@ -1,15 +1,12 @@
-using System;
-using System.Collections.Generic;
-using System.Threading.Tasks;
 using LLama.Common;
 using LLama.Native;
 using Microsoft.Extensions.Logging.Abstractions;
-using Xunit;
 
 namespace LLama.Unittest;
 
 [Trait("Category", "NoCI")]
-public class MtmdExecutorTests : IDisposable
+public sealed class MtmdExecutorTests
+    : IDisposable
 {
     private readonly LLamaWeights _weights;
     private readonly MtmdContextParams _mtmdParams;
@@ -49,9 +46,9 @@ public async Task InteractiveExecutor_EvaluateChunks_DoesNotRetokenize()
 
         executor.Embeds.Add(_mtmd.LoadMedia(Constants.MtmdImage));
 
-        await foreach (var _ in executor.InferAsync(prompt, new InferenceParams { MaxTokens = 0 }))
+        await foreach (var _ in executor.InferAsync(prompt, new InferenceParams { MaxTokens = 0 }, TestContext.Current.CancellationToken))
         {
-            Assert.True(false, "Prefill should not emit generated text");
+            Assert.Fail("Prefill should not emit generated text");
         }
 
         var diagnostics = executor.GetDiagnostics();
@@ -69,7 +66,7 @@ public async Task InstructExecutor_MtmdPromptAdvancesPastTokensOnce()
 
         var prompt = $"{_mtmdParams.MediaMarker ?? NativeApi.MtmdDefaultMarker() ?? "<media>"} Provide details.";
 
-        await foreach (var _ in executor.InferAsync(prompt, new InferenceParams { MaxTokens = 0 }))
+        await foreach (var _ in executor.InferAsync(prompt, new InferenceParams { MaxTokens = 0 }, TestContext.Current.CancellationToken))
         {
         }
 
diff --git a/LLama.Unittest/SamplingTests.cs b/LLama.Unittest/SamplingTests.cs
index 297641df3..fd921a011 100644
--- a/LLama.Unittest/SamplingTests.cs
+++ b/LLama.Unittest/SamplingTests.cs
@@ -3,8 +3,6 @@
 using System.Numerics.Tensors;
 using System.Text;
 
-using Xunit.Abstractions;
-
 namespace LLama.Unittest
 {
     public class SamplingTests
diff --git a/LLama.Unittest/StatelessExecutorTest.cs b/LLama.Unittest/StatelessExecutorTest.cs
index 1fae35a79..15ed924c8 100644
--- a/LLama.Unittest/StatelessExecutorTest.cs
+++ b/LLama.Unittest/StatelessExecutorTest.cs
@@ -1,7 +1,6 @@
 using System.Diagnostics;
 using LLama.Common;
 using LLama.Sampling;
-using Xunit.Abstractions;
 
 namespace LLama.Unittest
 {
@@ -43,8 +42,8 @@ public async Task Stateless()
             var timer = new Stopwatch();
             timer.Start();
 
-            var result1 = string.Join("", await executor.InferAsync(question, @params).ToListAsync());
-            var result2 = string.Join("", await executor.InferAsync(question, @params).ToListAsync());
+            var result1 = string.Join("", await executor.InferAsync(question, @params, TestContext.Current.CancellationToken).ToListAsync(TestContext.Current.CancellationToken));
+            var result2 = string.Join("", await executor.InferAsync(question, @params, TestContext.Current.CancellationToken).ToListAsync(TestContext.Current.CancellationToken));
 
             timer.Stop();
             _testOutputHelper.WriteLine($"{timer.ElapsedMilliseconds}ms");
@@ -74,8 +73,8 @@ public async Task OutOfContext_WithTruncateStrategy_SuccessfullyGenerates()
                 ContextTruncationPercentage = 0.2f // Drop 20% of tokens when full
             };
 
-            var result1 = string.Join("", await executor.InferAsync(question, @params).ToListAsync());
-            var result2 = string.Join("", await executor.InferAsync(question, @params).ToListAsync());
+            var result1 = string.Join("", await executor.InferAsync(question, @params, TestContext.Current.CancellationToken).ToListAsync(TestContext.Current.CancellationToken));
+            var result2 = string.Join("", await executor.InferAsync(question, @params, TestContext.Current.CancellationToken).ToListAsync(TestContext.Current.CancellationToken));
 
             _testOutputHelper.WriteLine(result1);
 
@@ -120,7 +119,7 @@ public async Task OutOfContext_WithDefaultStrategy_ThrowsException()
 
             var exception = await Assert.ThrowsAsync<Exceptions.ContextOverflowException>(async () =>
             {
-                await executor.InferAsync(question, @params).ToListAsync();
+                await executor.InferAsync(question, @params, TestContext.Current.CancellationToken).ToListAsync(TestContext.Current.CancellationToken);
             });
 
             _testOutputHelper.WriteLine($"Successfully caught expected exception: {exception.Message}");
@@ -155,7 +154,7 @@ public async Task OutOfContext_WithDefaultStrategy_2_ThrowsException()
 
             var exception = await Assert.ThrowsAsync<Exceptions.ContextOverflowException>(async () =>
             {
-                await executor.InferAsync(instruction, @params).ToListAsync();
+                await executor.InferAsync(instruction, @params, TestContext.Current.CancellationToken).ToListAsync(TestContext.Current.CancellationToken);
             });
 
             _testOutputHelper.WriteLine($"Successfully caught expected exception in InstructExecutor: {exception.Message}");
diff --git a/LLama.Unittest/StreamingTextDecoderTests.cs b/LLama.Unittest/StreamingTextDecoderTests.cs
index a53383207..3901a8fa1 100644
--- a/LLama.Unittest/StreamingTextDecoderTests.cs
+++ b/LLama.Unittest/StreamingTextDecoderTests.cs
@@ -1,6 +1,5 @@
 using System.Text;
 using LLama.Common;
-using Xunit.Abstractions;
 
 namespace LLama.Unittest;
 
diff --git a/LLama.Unittest/TemplateTests.cs b/LLama.Unittest/TemplateTests.cs
index 7579715a2..8ebef5a5f 100644
--- a/LLama.Unittest/TemplateTests.cs
+++ b/LLama.Unittest/TemplateTests.cs
@@ -1,8 +1,5 @@
 using System.Text;
 using LLama.Common;
-using LLama.Extensions;
-using LLama.Native;
-using Xunit.Abstractions;
 
 namespace LLama.Unittest;
 
diff --git a/LLama.Unittest/TextTransformTests.cs b/LLama.Unittest/TextTransformTests.cs
index 651ad4d06..fe3029d9f 100644
--- a/LLama.Unittest/TextTransformTests.cs
+++ b/LLama.Unittest/TextTransformTests.cs
@@ -1,4 +1,4 @@
-namespace LLama.Unittest
+namespace LLama.Unittest
 {
     public sealed class TextTransformTests
     {
@@ -21,7 +21,10 @@ public async Task EmptyTextOutputStreamTransformDoesNothing()
 
             var transform = new LLamaTransforms.EmptyTextOutputStreamTransform();
 
-            Assert.Equal(input, await transform.TransformAsync(input.ToAsyncEnumerable()).ToArrayAsync());
+            Assert.Equal(
+                input,
+                await transform.TransformAsync(input.ToAsyncEnumerable()).ToArrayAsync(TestContext.Current.CancellationToken)
+            );
         }
     }
 }
diff --git a/LLama.Unittest/TokenTests.cs b/LLama.Unittest/TokenTests.cs
index c159a0759..de965d8b5 100644
--- a/LLama.Unittest/TokenTests.cs
+++ b/LLama.Unittest/TokenTests.cs
@@ -30,12 +30,14 @@ public void TokensEndWith()
     {
         var tokens = _model.NativeHandle.Tokenize("The cat sat on the edge of the mat", false, true, Encoding.UTF8);
 
+#pragma warning disable CS0618 // Type or member is obsolete
         var result = tokens.TokensEndsWithAnyString(new[]
         {
             "a fish",
             "the mat",
             "this is an improbably long query to be using for this method"
         }, _model.NativeHandle, Encoding.UTF8);
+#pragma warning restore CS0618 // Type or member is obsolete
         Assert.True(result);
     }
 
@@ -44,10 +46,12 @@ public void TokensEndSubstring()
     {
         var tokens = _model.NativeHandle.Tokenize("The cat sat on the edge of the mat", false, true, Encoding.UTF8);
 
+#pragma warning disable CS0618 // Type or member is obsolete
         var result = tokens.TokensEndsWithAnyString((IList<string>)new[]
         {
             "at",
         }, _model.NativeHandle, Encoding.UTF8);
+#pragma warning restore CS0618 // Type or member is obsolete
         Assert.True(result);
     }
 
@@ -56,12 +60,14 @@ public void TokensNotEndWith()
     {
         var tokens = _model.NativeHandle.Tokenize("The cat sat on the edge of the mat", false, true, Encoding.UTF8);
 
+#pragma warning disable CS0618 // Type or member is obsolete
         var result = tokens.TokensEndsWithAnyString((IList<string>)new[]
         {
             "a fish",
             "The cat sat on the edge of the ma",
             "this is an improbably long query to be using for this method"
         }, _model.NativeHandle, Encoding.UTF8);
+#pragma warning restore CS0618 // Type or member is obsolete
         Assert.False(result);
     }
 
@@ -70,7 +76,9 @@ public void TokensNotEndWithNothing()
     {
         var tokens = _model.NativeHandle.Tokenize("The cat sat on the edge of the mat", false, true, Encoding.UTF8);
 
+#pragma warning disable CS0618 // Type or member is obsolete
         var result = tokens.TokensEndsWithAnyString((IList<string>)Array.Empty<string>(), _model.NativeHandle, Encoding.UTF8);
+#pragma warning restore CS0618 // Type or member is obsolete
         Assert.False(result);
     }
 
@@ -158,7 +166,9 @@ public void RoundTrip()
             var tokens = _model.NativeHandle.Tokenize(input, false, false, Encoding.UTF8);
 
             // Convert tokens back into characters
+#pragma warning disable CS0618 // Type or member is obsolete
             var chars = _model.NativeHandle.TokensToSpan(tokens, charsArr.AsSpan(), Encoding.UTF8);
+#pragma warning restore CS0618 // Type or member is obsolete
 
             // llama.cpp adds a space to the start of strings, remove that
             var output = new string(chars).TrimStart(' ');
diff --git a/LLama/Extensions/IModelParamsExtensions.cs b/LLama/Extensions/IModelParamsExtensions.cs
index f5753b389..2803f6655 100644
--- a/LLama/Extensions/IModelParamsExtensions.cs
+++ b/LLama/Extensions/IModelParamsExtensions.cs
@@ -131,7 +131,7 @@ private static IReadOnlyDictionary<string, IntPtr> GetAvailableBufferTypes()
             if (string.IsNullOrEmpty(name))
                 continue;
 
-            result[name] = buft;
+            result[name!] = buft;
         }
 
         return result;
diff --git a/LLama/Extensions/IntPtrExtensions.cs b/LLama/Extensions/IntPtrExtensions.cs
index eb5c90850..9701e50ab 100644
--- a/LLama/Extensions/IntPtrExtensions.cs
+++ b/LLama/Extensions/IntPtrExtensions.cs
@@ -1,9 +1,11 @@
 using System;
-using System.Runtime.InteropServices;
 using System.Text;
 
 namespace LLama.Extensions;
 
+/// <summary>
+/// Extensions for <see cref="IntPtr"/>
+/// </summary>
 public static class IntPtrExtensions
 {
 
diff --git a/LLama/LLamaInstructExecutor.cs b/LLama/LLamaInstructExecutor.cs
index e77f55b6f..a24210fb6 100644
--- a/LLama/LLamaInstructExecutor.cs
+++ b/LLama/LLamaInstructExecutor.cs
@@ -29,7 +29,7 @@ public class InstructExecutor
         private readonly string _instructionSuffix;
 
         /// <summary>
-        /// 
+        /// Construct a new InstructExecutor
         /// </summary>
         /// <param name="context"></param>
         /// <param name="instructionPrefix"></param>
@@ -47,6 +47,14 @@ public InstructExecutor(LLamaContext context,
             _instructionSuffix = instructionSuffix;
         }
 
+        /// <summary>
+        /// Construct a new InstructExecutor, with multimodal model weights
+        /// </summary>
+        /// <param name="context"></param>
+        /// <param name="clipModel"></param>
+        /// <param name="instructionPrefix"></param>
+        /// <param name="instructionSuffix"></param>
+        /// <param name="logger"></param>
         public InstructExecutor(LLamaContext context,
                                 MtmdWeights clipModel,
                                 string instructionPrefix = "\n\n### Instruction:\n\n",
diff --git a/LLama/LLamaInteractExecutor.cs b/LLama/LLamaInteractExecutor.cs
index 0fd512be7..9f8fc16fc 100644
--- a/LLama/LLamaInteractExecutor.cs
+++ b/LLama/LLamaInteractExecutor.cs
@@ -109,7 +109,7 @@ public override async Task LoadState(string filename, CancellationToken cancella
         {
             using var fs = new FileStream(filename, FileMode.Open, FileAccess.Read);
 
-            var state = await JsonSerializer.DeserializeAsync<InteractiveExecutorState>(fs);
+            var state = await JsonSerializer.DeserializeAsync<InteractiveExecutorState>(fs, cancellationToken: cancellationToken);
             await LoadState(state!, cancellationToken);
         }
 
@@ -117,8 +117,9 @@ public override async Task LoadState(string filename, CancellationToken cancella
         /// Decide whether generation should continue for the current iteration.
         /// </summary>
         /// <param name="args">Mutable inference state.</param>
+        /// <param name="cancellationToken"></param>
         /// <returns><c>true</c> to keep generating; otherwise <c>false</c>.</returns>
-        protected override Task<bool> GetLoopCondition(InferStateArgs args, CancellationToken cancellationToken)
+        protected override Task<bool> GetLoopCondition(InferStateArgs args, CancellationToken cancellationToken = default)
         {
             return Task.FromResult(args.RemainedTokens != 0 && !args.WaitForInput || _is_prompt_run);
         }
@@ -128,7 +129,8 @@ protected override Task<bool> GetLoopCondition(InferStateArgs args, Cancellation
         /// </summary>
         /// <param name="text">Prompt text or continuation provided by the caller.</param>
         /// <param name="args">Mutable inference state.</param>
-        protected override Task PreprocessInputs(string? text, InferStateArgs args, CancellationToken cancellationToken)
+        /// <param name="cancellationToken"></param>
+        protected override Task PreprocessInputs(string? text, InferStateArgs args, CancellationToken cancellationToken = default)
         {
             if (_is_prompt_run)
             {
diff --git a/LLama/Native/NativeApi.cs b/LLama/Native/NativeApi.cs
index 26b3c9dcd..3c58b754f 100644
--- a/LLama/Native/NativeApi.cs
+++ b/LLama/Native/NativeApi.cs
@@ -33,6 +33,10 @@ public static void llama_empty_call()
         [DllImport(libraryName, CallingConvention = CallingConvention.Cdecl)]
         public static extern long llama_max_devices();
 
+        /// <summary>
+        /// Get the maximum number of tensor buffer-type overrides supported by the framework
+        /// </summary>
+        /// <returns></returns>
         [DllImport(libraryName, CallingConvention = CallingConvention.Cdecl)]
         public static extern nint llama_max_tensor_buft_overrides();
 
@@ -472,6 +476,10 @@ public static string llama_split_prefix(string splitPath, int splitNo, int split
         [DllImport(ggmlBaseLibraryName, CallingConvention = CallingConvention.Cdecl)]
         public static extern IntPtr ggml_backend_buft_name(IntPtr buft);
 
+        /// <summary>
+        /// Get the current time in microseconds
+        /// </summary>
+        /// <returns></returns>
         [DllImport(libraryName, CallingConvention = CallingConvention.Cdecl)]
         public static extern long llama_time_us();
 
diff --git a/LLama/Native/SafeMtmdModelHandle.cs b/LLama/Native/SafeMtmdModelHandle.cs
index c5b541bcf..d4929953a 100644
--- a/LLama/Native/SafeMtmdModelHandle.cs
+++ b/LLama/Native/SafeMtmdModelHandle.cs
@@ -203,7 +203,7 @@ public int Tokenize(string text, bool addSpecial, bool parseSpecial, ReadOnlySpa
         /// <summary>
         /// Evaluate a batch of chunks using the helper (mirrors mtmd-helper eval logic).
         /// </summary>
-        /// <param name="chunks">Chunk collection produced by <see cref="Tokenize"/>.</param>
+        /// <param name="chunks">Chunk collection produced by <see cref="Tokenize(string,bool,bool,out LLama.Native.SafeMtmdInputChunks?)"/>.</param>
         /// <param name="llamaContext">Context handle that receives the evaluated tokens.</param>
         /// <param name="nPast">Number of past tokens; updated when evaluation succeeds.</param>
         /// <param name="seqId">Sequence identifier used for KV cache management.</param>

From c12adc4ba8e49841221dfd3f0410a431e2c6c3c8 Mon Sep 17 00:00:00 2001
From: Martin Evans <martindevans@gmail.com>
Date: Wed, 14 Oct 2026 23:59:35 +0100
Subject: [PATCH 3/5] Removed SpeechChat.cs, this allowed dropping NAudio and
 Whisper as dependencies.

---
 LLama.Examples/Examples/SpeechChat.cs | 240 --------------------------
 LLama.Examples/LLama.Examples.csproj  |   6 -
 2 files changed, 246 deletions(-)
 delete mode 100644 LLama.Examples/Examples/SpeechChat.cs

diff --git a/LLama.Examples/Examples/SpeechChat.cs b/LLama.Examples/Examples/SpeechChat.cs
deleted file mode 100644
index 596970eae..000000000
--- a/LLama.Examples/Examples/SpeechChat.cs
+++ /dev/null
@@ -1,240 +0,0 @@
-//using LLama.Common;
-//using NAudio.Wave;
-//using Spectre.Console;
-//using Whisper.net;
-
-//namespace LLama.Examples.Examples
-//{
-//    public static class SpeechChat
-//    {
-//        public static async Task Run()
-//        {
-//            AnsiConsole.MarkupLine("""
-//                                  [yellow on black]
-//                                  This example demonstrates the basics of audio transcriptions, speech recognition, and speech commands,
-//                                      as well as how to recognize a user's voice in real time and then get a response from LLM.
-//                                  It uses whisper.net and models could be found in: https://huggingface.co/ggerganov/whisper.cpp/tree/main.
-//                                  To use it, you need a working microphone and enough RAM to host both audio + language models.
-//                                  Once you've selected the models, just speak to your microphone and watch the LLM continue your text.
-//                                  While it's going, you can say something like 'Okay, stop', or 'Stop now', to interrupt the LLM's inference.
-                                  
-//                                  NOTE: You may need to poke around with the voice detection threshold, based on your mic's sensitivity.
-//                                  [/]
-//                                  """);
-
-//            AnsiConsole.MarkupLine("[white on black]You can find the official ggml models in whisper.cpp's huggingface repository: https://huggingface.co/ggerganov/whisper.cpp/tree/main [/]");
-
-//            var whisperModel = UserSettings.GetWhisperPath();
-//            var languageModel = UserSettings.GetModelPath();
-
-//            using var speechRecognitionServer = new SpeechRecognitionServer(whisperModel);
-
-//            using var _ = new LlamaSessionSpeechListener(speechRecognitionServer, languageModel);
-
-//            AnsiConsole.MarkupLine("[green]Voice active. Begin talking to transcribe. Press any key at any time to exit.[/]");
-//            await Task.Delay(1000);
-//            Console.ReadKey();
-//        }
-
-
-//        private class LlamaSessionSpeechListener
-//            : ISpeechListener, IDisposable
-//        {
-//            private bool _isModelResponding;
-//            private readonly SpeechRecognitionServer _audioServer;
-
-//            private readonly LLamaWeights _model;
-//            private readonly LLamaContext _context;
-//            private readonly InteractiveExecutor _executor;
-
-//            private string _fullPrompt = "";
-//            private bool _canceled;
-
-//            public LlamaSessionSpeechListener(SpeechRecognitionServer server, string languageModelPath)
-//            {
-//                var parameters = new ModelParams(languageModelPath)
-//                {
-//                    GpuLayerCount = 99
-//                };
-//                _model = LLamaWeights.LoadFromFile(parameters);
-//                _context = _model.CreateContext(parameters);
-//                _executor = new InteractiveExecutor(_context);
-
-//                _audioServer = server;
-//                _audioServer.ServiceUsers.Add(this);
-//            }
-
-//            // Whisper is struggling with single words and very short phrases without context, so it's actually better to say something like "Ok, Stop!" to have it work better.
-//            bool ISpeechListener.IsInterested(string audioTranscription) => !_isModelResponding || audioTranscription.Contains("stop", StringComparison.CurrentCultureIgnoreCase);
-//            void ISpeechListener.HandleSpeech(string audioTranscription)
-//            {
-//                if (_isModelResponding && audioTranscription.Contains("stop", StringComparison.CurrentCultureIgnoreCase))
-//                    _canceled = true;
-//                else if (!_isModelResponding)
-//                    _ = SendMessage(audioTranscription);
-//            }
-
-//            private async Task SendMessage(string newMessage)
-//            {
-//                // While a response is queried, we want to detect short phrases/commands like 'stop',
-//                _audioServer.DetectionSettings = (1, 2); // ..so we lower the min Speech Detection time.
-
-//                _isModelResponding = true;
-//                AddToPrompt($"\n{newMessage}\n", ConsoleColor.Blue);
-//                await foreach (var token in _executor.InferAsync(_fullPrompt))
-//                {
-//                    AddToPrompt(token);
-//                    if (_canceled)
-//                    {
-//                        AddToPrompt("[...stopped]", ConsoleColor.Red);
-//                        break;
-//                    }
-//                }
-//                _audioServer.DetectionSettings = (2, 3);         // Reset back to default detection settings to avoid false positives.
-//                (_isModelResponding, _canceled) = (false, false); // Reset the state variables to their default.
-//            }
-
-//            private void AddToPrompt(string msg, ConsoleColor color = ConsoleColor.Yellow)
-//            {
-//                _fullPrompt += msg;
-
-//                AnsiConsole.Markup($"[{color}]{Markup.Escape(msg)}[/]");
-//            }
-
-//            void IDisposable.Dispose()
-//            {
-//                _model.Dispose();
-//                _context.Dispose();
-//            }
-//        }
-
-//        public interface ISpeechListener
-//        {
-//            bool IsInterested(string audioTranscription);
-//            void HandleSpeech(string audioTranscription);
-//        }
-
-//        public sealed class SpeechRecognitionServer
-//            : IDisposable
-//        {
-//            private static readonly TimeSpan ClipLength = TimeSpan.FromMilliseconds(250);
-
-//            private const float VoiceDetectionThreshold = 0.01f; // Adjust as needed
-//            private readonly string[] _knownFalsePositives = ["[BLANK_AUDIO]", "Thank you", "[silence]"];
-
-//            private readonly WaveInEvent _waveIn;
-
-//            private readonly WaveFormat _waveFormat = new(16000, 16, 1); // 16KHz, 16 bits, Mono Channel
-//            private readonly List<byte> _recordedBytes = [];
-
-//            private readonly WhisperProcessor? _processor;
-
-//            private readonly string _whisperPrompt =
-//"""
-//The short audio comes from a user that is speaking to an AI Language Model in real time.
-//Pay extra attentions for commands like 'ok stop' or just 'stop'.
-//In case of inaudible sentences that might be, assume they're saying 'stop'.
-//""".Trim();
-
-//            // Tracked stats for Speech Recognition, Parsing, and Serving.
-//            private int _currentBlankClips;  // Ideally would work with milliseconds,
-//            private int _totalNonBlankClips; // ..but for example's sake they work on a
-
-//            private int _nonIdleTime;        // ..clip-based quant-length (1 = clipLength).
-//            // Default detection settings: A speech of 750ms, followed by pause of 500ms. (2x250ms)
-//            public (int minBlanksPerSeparation, int minNonBlanksForValidMessages) DetectionSettings = (2, 3);
-
-//            public readonly HashSet<ISpeechListener> ServiceUsers = [];
-
-//            public SpeechRecognitionServer(string modelPath)
-//            {
-//                var whisperFactory = WhisperFactory.FromPath(modelPath);
-
-//                var builder = whisperFactory
-//                    .CreateBuilder()
-//                    .WithThreads(16)
-//                    .WithPrompt(_whisperPrompt)
-//                    .WithSingleSegment()
-//                    .WithLanguage("en");
-//                ((BeamSearchSamplingStrategyBuilder)builder.WithBeamSearchSamplingStrategy()).WithPatience(0.2f).WithBeamSize(5);
-//                _processor = builder.Build();
-
-//                _waveIn = new WaveInEvent { BufferMilliseconds = (int)ClipLength.TotalMilliseconds, WaveFormat = _waveFormat };
-//                _waveIn.DataAvailable += OnAudioDataAvailable;
-//                _waveIn.StartRecording();
-//            }
-
-//            private void OnAudioDataAvailable(object? sender, WaveInEventArgs e)
-//            {
-//                // Cache the recorded bytes
-//                _recordedBytes.AddRange(e.Buffer[..e.BytesRecorded]);
-//                if (_recordedBytes.Count > 110000000) { _recordedBytes.RemoveRange(0, 50000000); }
-
-//                // Get the max volume contained inside the clip. Since the clip is recorded as bytes, we need to translate them to samples before getting their volume.
-//                var maxVolume = 0f; // This byte->sample algorithm is from: https://github.com/naudio/NAudio/blob/master/Docs/RecordingLevelMeter.md#calculating-peak-values
-//                for (var i = 0; i < e.BytesRecorded; i += 2)
-//                {
-//                    maxVolume = Math.Max(maxVolume, Math.Abs((short)((e.Buffer[i + 1] << 8) | e.Buffer[i + 0]) / 32768f));
-//                }
-
-//                // Compare the volume with the threshold and act accordingly. Once an interesting and 'full' set of clips pops up, serve it.
-//                if (maxVolume >= VoiceDetectionThreshold)
-//                {
-//                    _currentBlankClips = 0;
-//                    _totalNonBlankClips++;
-//                    _nonIdleTime++;
-//                }
-//                else if (++_currentBlankClips < DetectionSettings.minBlanksPerSeparation)
-//                {
-//                    _nonIdleTime++;
-//                }
-//                else
-//                {
-//                    if (_totalNonBlankClips >= DetectionSettings.minNonBlanksForValidMessages)
-//                        SendTranscription();
-//                    else if (_totalNonBlankClips > 0) { } // This might be case of a false-positive -- knock, noise, cough, anything.
-//                    (_currentBlankClips, _totalNonBlankClips, _nonIdleTime) = (0, 0, 0);
-//                }
-
-
-//                async void SendTranscription()
-//                {
-//                    var bytesPerClip = _waveFormat.BitsPerSample * (int)ClipLength.TotalMilliseconds * 2;
-//                    var capturedClipBytes = _recordedBytes.TakeLast(bytesPerClip * (_nonIdleTime + 2)).ToArray();
-
-//                    // Save to temporary file.
-//                    var transcribedText = await ProcessAudio(capturedClipBytes, "Assets\\temp.wav"); 
-
-//                    // False positive.. yikes!
-//                    if (_knownFalsePositives.Contains(transcribedText))
-//                        return;    
-                    
-//                    foreach (var user in ServiceUsers.Where(x => x.IsInterested(transcribedText)))
-//                        user.HandleSpeech(transcribedText);
-//                }
-//            }
-
-//            /// <summary> Requests a transcription and responds with the text. </summary>
-//            private async Task<string> ProcessAudio(byte[] bytes, string tempWavFilePath)
-//            {
-//                await using var wavStream = new MemoryStream();
-
-//                await using (var writer = new WaveFileWriter(tempWavFilePath, _waveFormat))
-//                    writer.Write(bytes, 0, bytes.Length);
-//                await using (var fileStream = File.OpenRead(tempWavFilePath))
-//                    await fileStream.CopyToAsync(wavStream);
-
-//                wavStream.Seek(0, SeekOrigin.Begin);
-
-//                Console.Beep();
-//                return string.Join(' ', await _processor!.ProcessAsync(wavStream).Select(x => x.Text).ToListAsync()).Trim();
-//            }
-
-//            void IDisposable.Dispose()
-//            {
-//                _waveIn.Dispose();
-//                _processor?.Dispose();
-//            }
-//        }
-//    }
-//}
diff --git a/LLama.Examples/LLama.Examples.csproj b/LLama.Examples/LLama.Examples.csproj
index 32f55cb63..0963dc3f7 100644
--- a/LLama.Examples/LLama.Examples.csproj
+++ b/LLama.Examples/LLama.Examples.csproj
@@ -15,13 +15,7 @@
 
   <ItemGroup>
     <PackageReference Include="Microsoft.Extensions.Logging.Console" Version="10.0.13" />
-    <PackageReference Include="NAudio" Version="3.1.0" />
     <PackageReference Include="Spectre.Console" Version="0.57.2" />
-    <PackageReference Include="Whisper.net" Version="1.9.1" />
-    <PackageReference Include="Whisper.net.Runtime" Version="1.9.1" />
-    <PackageReference Include="Whisper.net.Runtime.Clblast" Version="1.5.0" />
-    <PackageReference Include="Whisper.net.Runtime.CoreML" Version="1.9.1" />
-    <PackageReference Include="Whisper.net.Runtime.Cublas" Version="1.5.0" />
     <PackageReference Include="Microsoft.Agents.AI" Version="*-*" />
     <PackageReference Include="Microsoft.Extensions.AI" Version="*-*" />
     <PackageReference Include="CommunityToolkit.VectorData.SqliteVec" Version="*-*" />

From 8da20044ac5393ad4a47ed723c4abeb4e862ce88 Mon Sep 17 00:00:00 2001
From: Martin Evans <martindevans@gmail.com>
Date: Thu, 15 Oct 2026 00:02:28 +0100
Subject: [PATCH 4/5] Updated CI from dotnet 8.0 to dotnet 10.0

---
 .github/workflows/main.yml | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/.github/workflows/main.yml b/.github/workflows/main.yml
index 419fd3eaa..e72c8496e 100644
--- a/.github/workflows/main.yml
+++ b/.github/workflows/main.yml
@@ -52,12 +52,12 @@ jobs:
         apk add --no-cache git bash icu-libs libstdc++ libgomp
         dotnet --list-runtimes
         dotnet --list-sdks
-    - name: Install .NET 8.0 runtime (Alpine)
+    - name: Install .NET 10.0 runtime (Alpine)
       if: ${{ matrix.build == 'linux-musl-x64-cpu' }}
       run: |
         wget https://dot.net/v1/dotnet-install.sh -O dotnet-install.sh
         chmod +x dotnet-install.sh
-        ./dotnet-install.sh --runtime dotnet --channel 8.0 --install-dir /usr/share/dotnet
+        ./dotnet-install.sh --runtime dotnet --channel 10.0 --install-dir /usr/share/dotnet
         rm dotnet-install.sh
 
     - uses: actions/checkout@v6
@@ -65,7 +65,7 @@ jobs:
       if: ${{ matrix.build != 'linux-musl-x64-cpu' }}
       with:
         dotnet-version: |
-          8.0.x
+          10.0.x
     - name: Install Mobile Workloads
       if: ${{ contains(runner.os, 'windows') }}
       run: |

From 4a5b46ebfd3252430dbffb858653824aa17f93c2 Mon Sep 17 00:00:00 2001
From: Martin Evans <martindevans@gmail.com>
Date: Thu, 15 Oct 2026 00:03:42 +0100
Subject: [PATCH 5/5] Cleaned up unused `using` statements

---
 LLama/Common/ContextOverflowStrategy.cs  | 4 ----
 LLama/Common/InferenceParams.cs          | 2 --
 LLama/Extensions/DictionaryExtensions.cs | 1 -
 LLama/LLamaInstructExecutor.cs           | 1 -
 LLama/LLamaInteractExecutor.cs           | 2 --
 LLama/LLamaWeights.cs                    | 1 -
 LLama/Native/SafeLLamaSamplerHandle.cs   | 2 --
 LLama/StreamingTokenDecoder.cs           | 2 --
 8 files changed, 15 deletions(-)

diff --git a/LLama/Common/ContextOverflowStrategy.cs b/LLama/Common/ContextOverflowStrategy.cs
index cf52b8bd0..7dd755921 100644
--- a/LLama/Common/ContextOverflowStrategy.cs
+++ b/LLama/Common/ContextOverflowStrategy.cs
@@ -1,7 +1,3 @@
-using System;
-using System.Collections.Generic;
-using System.Text;
-
 namespace LLama.Common
 {
     /// <summary>
diff --git a/LLama/Common/InferenceParams.cs b/LLama/Common/InferenceParams.cs
index 37cbeb034..dbacf941c 100644
--- a/LLama/Common/InferenceParams.cs
+++ b/LLama/Common/InferenceParams.cs
@@ -1,8 +1,6 @@
 using LLama.Abstractions;
 using System.Collections.Generic;
-using LLama.Native;
 using LLama.Sampling;
-using System;
 
 namespace LLama.Common
 {
diff --git a/LLama/Extensions/DictionaryExtensions.cs b/LLama/Extensions/DictionaryExtensions.cs
index aaa782eae..ce56980dc 100644
--- a/LLama/Extensions/DictionaryExtensions.cs
+++ b/LLama/Extensions/DictionaryExtensions.cs
@@ -1,5 +1,4 @@
 using System.Collections.Generic;
-using LLama.Native;
 
 namespace LLama.Extensions
 {
diff --git a/LLama/LLamaInstructExecutor.cs b/LLama/LLamaInstructExecutor.cs
index a24210fb6..c7aee4851 100644
--- a/LLama/LLamaInstructExecutor.cs
+++ b/LLama/LLamaInstructExecutor.cs
@@ -11,7 +11,6 @@
 using LLama.Common;
 using LLama.Exceptions;
 using LLama.Native;
-using LLama.Sampling;
 using Microsoft.Extensions.Logging;
 
 namespace LLama
diff --git a/LLama/LLamaInteractExecutor.cs b/LLama/LLamaInteractExecutor.cs
index 9f8fc16fc..104758dbd 100644
--- a/LLama/LLamaInteractExecutor.cs
+++ b/LLama/LLamaInteractExecutor.cs
@@ -8,10 +8,8 @@
 using System.Threading.Tasks;
 using LLama.Abstractions;
 using LLama.Common;
-using LLama;
 using LLama.Exceptions;
 using LLama.Native;
-using LLama.Sampling;
 using Microsoft.Extensions.Logging;
 
 
diff --git a/LLama/LLamaWeights.cs b/LLama/LLamaWeights.cs
index 50098b6b3..9ad9b9c0b 100644
--- a/LLama/LLamaWeights.cs
+++ b/LLama/LLamaWeights.cs
@@ -1,6 +1,5 @@
 using System;
 using System.Collections.Generic;
-using System.Linq;
 using System.Text;
 using System.Threading;
 using System.Threading.Tasks;
diff --git a/LLama/Native/SafeLLamaSamplerHandle.cs b/LLama/Native/SafeLLamaSamplerHandle.cs
index 9de5eada7..447c51bee 100644
--- a/LLama/Native/SafeLLamaSamplerHandle.cs
+++ b/LLama/Native/SafeLLamaSamplerHandle.cs
@@ -1,8 +1,6 @@
 using System;
 using System.Collections.Generic;
-using System.Runtime.InteropServices;
 using System.Text;
-using LLama.Extensions;
 
 namespace LLama.Native;
 
diff --git a/LLama/StreamingTokenDecoder.cs b/LLama/StreamingTokenDecoder.cs
index cd821fb3b..b71eec90c 100644
--- a/LLama/StreamingTokenDecoder.cs
+++ b/LLama/StreamingTokenDecoder.cs
@@ -1,8 +1,6 @@
 using System.Diagnostics;
 using System;
 using System.Collections.Generic;
-using System.Linq;
-using System.Runtime.InteropServices;
 using System.Text;
 using LLama.Native;