Merge pull request #672 from SignalRT/master

Reintroduce MacOS on CI

.github/workflows/main.yml

@@ -15,20 +15,22 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        build: [linux-release, windows-release]
+        build: [linux-release, windows-release, osx-release]
         include:
           - build: linux-release
             os: ubuntu-latest
             config: release
+          - build: osx-release  
+            os: macos-14 # https://github.blog/changelog/2024-01-30-github-actions-introducing-the-new-m1-macos-runner-available-to-open-source/
+            config: release            
           - build: windows-release
             os: windows-2019
             config: release
     steps:
     - uses: actions/checkout@v4
     - uses: actions/setup-dotnet@v4
       with:
-        dotnet-version: | 
-          7.0.x
+        dotnet-version: |
           8.0.x
     - name: Cache Packages
       uses: actions/cache@v4
@@ -43,7 +45,7 @@ jobs:
     - name: Build
       run: dotnet build LLamaSharp.sln -c ${{ matrix.config }} --no-restore
     - name: Test
-      run: dotnet test LLamaSharp.sln -c ${{ matrix.config }} -l "console;verbosity=detailed" --diag:logs/log.txt
+      run: dotnet test LLamaSharp.sln -c ${{ matrix.config }} -l "console;verbosity=detailed" --diag:logs/log.txt --filter Category!=NoCI
     - name: Upload artifacts
       if: always()
       uses: actions/upload-artifact@v3

LLama.Unittest/BasicTest.cs

@@ -17,7 +17,8 @@ public BasicTest(ITestOutputHelper testOutputHelper)
             _testOutputHelper = testOutputHelper;
             _params = new ModelParams(Constants.GenerativeModelPath)
             {
-                ContextSize = 2048
+                ContextSize = 2048,
+                GpuLayerCount = Constants.CIGpuLayerCount
             };
             _model = LLamaWeights.LoadFromFile(_params);
         }

LLama.Unittest/BeamTests.cs

@@ -17,7 +17,8 @@ public BeamTests(ITestOutputHelper testOutputHelper)
         _testOutputHelper = testOutputHelper;
         _params = new ModelParams(Constants.GenerativeModelPath)
         {
-            ContextSize = 2048
+            ContextSize = 2048,
+            GpuLayerCount = Constants.CIGpuLayerCount,
         };
         _model = LLamaWeights.LoadFromFile(_params);
     }
@@ -27,7 +28,6 @@ public void Dispose()
         _model.Dispose();
     }
 
-    //[Fact(Skip = "Very very slow in CI")]
     [Fact]
     public void BasicBeam()
     {

LLama.Unittest/Constants.cs

@@ -1,4 +1,6 @@
-namespace LLama.Unittest
+using System.Runtime.InteropServices;
+
+namespace LLama.Unittest
 {
     internal static class Constants
     {
@@ -8,5 +10,25 @@ internal static class Constants
         public static readonly string LLavaModelPath = "Models/llava-v1.6-mistral-7b.Q3_K_XS.gguf";
         public static readonly string LLavaMmpPath = "Models/mmproj-model-f16.gguf";
         public static readonly string LLavaImage = "Models/extreme-ironing-taxi-610x427.jpg";
+
+        /// <summary>
+        /// Calculate GpuLayer Count to use in UnitTest
+        /// </summary>
+        /// <returns> Defaults to 20 in all the cases, except MacOS/OSX release (to disable METAL on github CI)</returns>
+        public static int CIGpuLayerCount
+        {
+            get
+            {
+                if (RuntimeInformation.IsOSPlatform(OSPlatform.OSX))
+                {
+                    #if DEBUG
+                      return 20;
+                    #else
+                      return 0;                      
+                    #endif
+                }
+                else return 20;
+            }
+        }
     }
 }

LLama.Unittest/GrammarTest.cs

@@ -16,6 +16,7 @@ public GrammarTest()
             {
                 ContextSize = 2048,
                 Seed = 92,
+                GpuLayerCount = Constants.CIGpuLayerCount,                
             };
             _model = LLamaWeights.LoadFromFile(_params);
         }

LLama.Unittest/LLamaContextTests.cs

@@ -14,6 +14,7 @@ public LLamaContextTests()
             var @params = new ModelParams(Constants.GenerativeModelPath)
             {
                 ContextSize = 768,
+                GpuLayerCount = Constants.CIGpuLayerCount,
             };
             _weights = LLamaWeights.LoadFromFile(@params);
             _context = _weights.CreateContext(@params);

LLama.Unittest/LLamaEmbedderTests.cs

@@ -1,4 +1,4 @@
-using LLama.Common;
+using LLama.Common;
 using Xunit.Abstractions;
 
 namespace LLama.Unittest;
@@ -10,7 +10,21 @@ public sealed class LLamaEmbedderTests
     public LLamaEmbedderTests(ITestOutputHelper testOutputHelper)
     {
         _testOutputHelper = testOutputHelper;
-
+
+        var @params = new ModelParams(Constants.EmbeddingModelPath)
+        {
+            ContextSize = 4096,
+            Threads = 5,
+            Embeddings = true,
+            GpuLayerCount = Constants.CIGpuLayerCount,
+        };
+        using var weights = LLamaWeights.LoadFromFile(@params);
+        _embedder = new(weights, @params);
+    }
+
+    public void Dispose()
+    {
+        _embedder.Dispose();
     }
 
     private static float Dot(float[] a, float[] b)

LLama.Unittest/LLavaWeightsTests.cs

@@ -17,7 +17,8 @@ public LLavaWeightTests()
             var @params = new ModelParams(Constants.GenerativeModelPath)
             {
                 // Llava models requires big context
-                ContextSize = 4096
+                ContextSize = 4096,
+                GpuLayerCount = Constants.CIGpuLayerCount,                
             };
             _llamaWeights = LLamaWeights.LoadFromFile(@params);
             _lLavaWeights = LLavaWeights.LoadFromFile(Constants.LLavaMmpPath);
@@ -32,15 +33,15 @@ public void Dispose()
             _lLavaWeights.Dispose();
         }
 
-        [Fact(Skip = "Very very slow in CI")]
+        [Fact,Trait("Category", "NoCI")]
         public void EmbedImageAsFileName()
         {
             int n_past = 0;
             SafeLlavaImageEmbedHandle emb = _lLavaWeights.CreateImageEmbeddings(_context, Constants.LLavaImage);
             Assert.True( _lLavaWeights.EvalImageEmbed( _context, emb, ref n_past ) );
         }        
 
-        [Fact(Skip = "Very very slow in CI")]
+        [Fact,Trait("Category", "NoCI")]
         public void EmbedImageAsBinary()
         {
             int n_past = 0;

LLama.Unittest/MemoryDisposalTests.cs

@@ -9,7 +9,8 @@ public void ModelDisposal()
     {
         var @params = new ModelParams(Constants.GenerativeModelPath)
         {
-            ContextSize = 2048
+            ContextSize = 2048,
+            GpuLayerCount = 0,
         };
         var model = LLamaWeights.LoadFromFile(@params);
 
@@ -23,7 +24,8 @@ public void ContextDisposal()
     {
         var @params = new ModelParams(Constants.GenerativeModelPath)
         {
-            ContextSize = 2048
+            ContextSize = 2048,
+            GpuLayerCount = Constants.CIGpuLayerCount,            
         };
         var model = LLamaWeights.LoadFromFile(@params);
 

LLama.Unittest/StatelessExecutorTest.cs

@@ -20,6 +20,7 @@ public StatelessExecutorTest(ITestOutputHelper testOutputHelper)
                 ContextSize = 60,
                 Seed = 1754,
                 BatchSize = 2,
+                GpuLayerCount = Constants.CIGpuLayerCount,                
             };
             _weights = LLamaWeights.LoadFromFile(_params);
         }

LLama.Unittest/TokenTests.cs

@@ -14,7 +14,8 @@ public TokenTests()
     {
         _params = new ModelParams(Constants.GenerativeModelPath)
         {
-            ContextSize = 2048
+            ContextSize = 2048,
+            GpuLayerCount = Constants.CIGpuLayerCount,
         };
         _model = LLamaWeights.LoadFromFile(_params);
     }