Merge pull request #1102 from stephentoub/updatemeai

Update Microsoft.Extensions.AI to 9.3.0-preview.1.25114.11

Author: martindevans

LLama.Unittest/LLamaEmbedderTests.cs

@@ -43,10 +43,10 @@ private async Task CompareEmbeddings(string modelPath)
         Assert.DoesNotContain(float.NaN, spoon);
 
         var generator = (IEmbeddingGenerator<string, Embedding<float>>)embedder;
-        Assert.NotNull(generator.Metadata);
-        Assert.Equal(nameof(LLamaEmbedder), generator.Metadata.ProviderName);
-        Assert.NotNull(generator.Metadata.ModelId);
-        Assert.NotEmpty(generator.Metadata.ModelId);
+        Assert.NotNull(generator.GetService<EmbeddingGeneratorMetadata>());
+        Assert.Equal(nameof(LLamaEmbedder), generator.GetService<EmbeddingGeneratorMetadata>()?.ProviderName);
+        Assert.NotNull(generator.GetService<EmbeddingGeneratorMetadata>()?.ModelId);
+        Assert.NotEmpty(generator.GetService<EmbeddingGeneratorMetadata>()?.ModelId!);
         Assert.Same(embedder, generator.GetService<LLamaEmbedder>());
         Assert.Same(generator, generator.GetService<IEmbeddingGenerator<string, Embedding<float>>>());
         Assert.Null(generator.GetService<string>());

LLama/Extensions/LLamaExecutorExtensions.cs

@@ -36,6 +36,7 @@ private sealed class LLamaExecutorChatClient(
         IHistoryTransform? historyTransform = null,
         ITextStreamTransform? outputTransform = null) : IChatClient
     {
+        private static readonly ChatClientMetadata s_metadata = new(nameof(LLamaExecutorChatClient));
         private static readonly InferenceParams s_defaultParams = new();
         private static readonly DefaultSamplingPipeline s_defaultPipeline = new();
         private static readonly string[] s_antiPrompts = ["User:", "Assistant:", "System:"];
@@ -47,21 +48,19 @@ private sealed class LLamaExecutorChatClient(
         private readonly ITextStreamTransform _outputTransform = outputTransform ??
             new LLamaTransforms.KeywordTextOutputStreamTransform(s_antiPrompts);
 
-        /// <inheritdoc/>
-        public ChatClientMetadata Metadata { get; } = new(nameof(LLamaExecutorChatClient));
-
         /// <inheritdoc/>
         public void Dispose() { }
 
         /// <inheritdoc/>
-        public object? GetService(Type serviceType, object? key = null) =>
-            key is not null ? null :
+        public object? GetService(Type serviceType, object? serviceKey = null) =>
+            serviceKey is not null ? null :
+            serviceType == typeof(ChatClientMetadata) ? s_metadata :
             serviceType?.IsInstanceOfType(_executor) is true ? _executor :
             serviceType?.IsInstanceOfType(this) is true ? this :
             null;
 
         /// <inheritdoc/>
-        public async Task<ChatCompletion> CompleteAsync(
+        public async Task<ChatResponse> GetResponseAsync(
             IList<ChatMessage> chatMessages, ChatOptions? options = null, CancellationToken cancellationToken = default)
         {
             var result = _executor.InferAsync(CreatePrompt(chatMessages), CreateInferenceParams(options), cancellationToken);
@@ -79,7 +78,7 @@ public async Task<ChatCompletion> CompleteAsync(
         }
 
         /// <inheritdoc/>
-        public async IAsyncEnumerable<StreamingChatCompletionUpdate> CompleteStreamingAsync(
+        public async IAsyncEnumerable<ChatResponseUpdate> GetStreamingResponseAsync(
             IList<ChatMessage> chatMessages, ChatOptions? options = null, [EnumeratorCancellation] CancellationToken cancellationToken = default)
         {
             var result = _executor.InferAsync(CreatePrompt(chatMessages), CreateInferenceParams(options), cancellationToken);
@@ -142,8 +141,8 @@ private string CreatePrompt(IList<ChatMessage> messages)
                 MaxTokens = options?.MaxOutputTokens ?? 256, // arbitrary upper limit
                 SamplingPipeline = new DefaultSamplingPipeline()
                 {
-                    FrequencyPenalty = options?.AdditionalProperties?.TryGetValue(nameof(DefaultSamplingPipeline.FrequencyPenalty), out float af) is true ? af : s_defaultPipeline.FrequencyPenalty,
-                    PresencePenalty = options?.AdditionalProperties?.TryGetValue(nameof(DefaultSamplingPipeline.PresencePenalty), out float ap) is true ? ap : s_defaultPipeline.PresencePenalty,
+                    FrequencyPenalty = options?.FrequencyPenalty ?? s_defaultPipeline.FrequencyPenalty,
+                    PresencePenalty = options?.PresencePenalty ?? s_defaultPipeline.PresencePenalty,
                     PreventEOS = options?.AdditionalProperties?.TryGetValue(nameof(DefaultSamplingPipeline.PreventEOS), out bool eos) is true ? eos : s_defaultPipeline.PreventEOS,
                     PenalizeNewline = options?.AdditionalProperties?.TryGetValue(nameof(DefaultSamplingPipeline.PenalizeNewline), out bool pnl) is true ? pnl : s_defaultPipeline.PenalizeNewline,
                     RepeatPenalty = options?.AdditionalProperties?.TryGetValue(nameof(DefaultSamplingPipeline.RepeatPenalty), out float rp) is true ? rp : s_defaultPipeline.RepeatPenalty,
@@ -152,8 +151,8 @@ private string CreatePrompt(IList<ChatMessage> messages)
                     MinKeep = options?.AdditionalProperties?.TryGetValue(nameof(DefaultSamplingPipeline.MinKeep), out int mk) is true ? mk : s_defaultPipeline.MinKeep,
                     MinP = options?.AdditionalProperties?.TryGetValue(nameof(DefaultSamplingPipeline.MinP), out float mp) is true ? mp : s_defaultPipeline.MinP,
                     Seed = options?.Seed is long seed ? (uint)seed : (uint)(t_random ??= new()).Next(),
-                    Temperature = options?.Temperature ?? 0,
-                    TopP = options?.TopP ?? 0,
+                    Temperature = options?.Temperature ?? s_defaultPipeline.Temperature,
+                    TopP = options?.TopP ?? s_defaultPipeline.TopP,
                     TopK = options?.TopK ?? s_defaultPipeline.TopK,
                     TypicalP = options?.AdditionalProperties?.TryGetValue(nameof(DefaultSamplingPipeline.TypicalP), out float tp) is true ? tp : s_defaultPipeline.TypicalP,
                 },

LLama/LLamaEmbedder.EmbeddingGenerator.cs

@@ -14,18 +14,31 @@ public partial class LLamaEmbedder
     private EmbeddingGeneratorMetadata? _metadata;
 
     /// <inheritdoc />
-    EmbeddingGeneratorMetadata IEmbeddingGenerator<string, Embedding<float>>.Metadata =>
-        _metadata ??= new(
-            nameof(LLamaEmbedder),
-            modelId: Context.NativeHandle.ModelHandle.ReadMetadata().TryGetValue("general.name", out var name) ? name : null,
-            dimensions: EmbeddingSize);
+    object? IEmbeddingGenerator<string, Embedding<float>>.GetService(Type serviceType, object? serviceKey)
+    {
+        if (serviceKey is null)
+        {
+            if (serviceType == typeof(EmbeddingGeneratorMetadata))
+            {
+                return _metadata ??= new(
+                    nameof(LLamaEmbedder),
+                    modelId: Context.NativeHandle.ModelHandle.ReadMetadata().TryGetValue("general.name", out var name) ? name : null,
+                    dimensions: EmbeddingSize);
+            }
 
-    /// <inheritdoc />
-    object? IEmbeddingGenerator<string, Embedding<float>>.GetService(Type serviceType, object? key) =>
-        key is not null ? null :
-        serviceType?.IsInstanceOfType(Context) is true ? Context :
-        serviceType?.IsInstanceOfType(this) is true ? this :
-        null;
+            if (serviceType?.IsInstanceOfType(Context) is true)
+            {
+                return Context;
+            }
+
+            if (serviceType?.IsInstanceOfType(this) is true)
+            {
+                return this;
+            }
+        }
+
+        return null;
+    }
 
     /// <inheritdoc />
     async Task<GeneratedEmbeddings<Embedding<float>>> IEmbeddingGenerator<string, Embedding<float>>.GenerateAsync(IEnumerable<string> values, EmbeddingGenerationOptions? options, CancellationToken cancellationToken)

LLama/LLamaSharp.csproj

@@ -50,7 +50,7 @@
 
   <ItemGroup>
     <PackageReference Include="Microsoft.Bcl.AsyncInterfaces" Version="9.0.0" />
-    <PackageReference Include="Microsoft.Extensions.AI.Abstractions" Version="9.1.0-preview.1.25064.3" />
+    <PackageReference Include="Microsoft.Extensions.AI.Abstractions" Version="9.3.0-preview.1.25114.11" />
     <PackageReference Include="Microsoft.Extensions.Logging.Abstractions" Version="9.0.0" />
     <PackageReference Include="System.Numerics.Tensors" Version="9.0.0" />
   </ItemGroup>