Set GPULayerCount to execute the Test

Set GPULayerCount to default value (20) to execute UnitTest. In the case of Release Execution on MacOS set the value to ZERO to disable METAL on MacOS and be able to execute it in CI.

LLama.Unittest/BasicTest.cs

@@ -18,7 +18,7 @@ namespace LLama.Unittest
             _params = new ModelParams(Constants.GenerativeModelPath)
             {
                 ContextSize = 2048,
-                GpuLayerCount = 0
+                GpuLayerCount = Constants.CIGpuLayerCount
             };
             _model = LLamaWeights.LoadFromFile(_params);
         }

LLama.Unittest/BeamTests.cs

@@ -18,7 +18,7 @@ public sealed class BeamTests
         _params = new ModelParams(Constants.GenerativeModelPath)
         {
             ContextSize = 2048,
-            GpuLayerCount = 0,
+            GpuLayerCount = Constants.CIGpuLayerCount,
         };
         _model = LLamaWeights.LoadFromFile(_params);
     }

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 @@
         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 IOS 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

@@ -16,7 +16,7 @@ namespace LLama.Unittest
             {
                 ContextSize = 2048,
                 Seed = 92,
-                GpuLayerCount = 0,                
+                GpuLayerCount = Constants.CIGpuLayerCount,                
             };
             _model = LLamaWeights.LoadFromFile(_params);
         }

LLama.Unittest/LLamaContextTests.cs

@@ -14,7 +14,7 @@ namespace LLama.Unittest
             var @params = new ModelParams(Constants.GenerativeModelPath)
             {
                 ContextSize = 768,
-                GpuLayerCount = 0
+                GpuLayerCount = Constants.CIGpuLayerCount,
             };
             _weights = LLamaWeights.LoadFromFile(@params);
             _context = _weights.CreateContext(@params);

LLama.Unittest/LLamaEmbedderTests.cs

@@ -19,7 +19,7 @@ public sealed class LLamaEmbedderTests
             ContextSize = 4096,
             Threads = 5,
             Embeddings = true,
-            GpuLayerCount = 0
+            GpuLayerCount = Constants.CIGpuLayerCount,
         };
         using var weights = LLamaWeights.LoadFromFile(@params);
         _embedder = new(weights, @params);

LLama.Unittest/LLavaWeightsTests.cs

@@ -18,7 +18,7 @@ namespace LLama.Unittest
             {
                 // Llava models requires big context
                 ContextSize = 4096,
-                GpuLayerCount = 0                
+                GpuLayerCount = Constants.CIGpuLayerCount,                
             };
             _llamaWeights = LLamaWeights.LoadFromFile(@params);
             _lLavaWeights = LLavaWeights.LoadFromFile(Constants.LLavaMmpPath);

LLama.Unittest/MemoryDisposalTests.cs

@@ -25,7 +25,7 @@ public class MemoryDisposalTests
         var @params = new ModelParams(Constants.GenerativeModelPath)
         {
             ContextSize = 2048,
-            GpuLayerCount = 0            
+            GpuLayerCount = Constants.CIGpuLayerCount,            
         };
         var model = LLamaWeights.LoadFromFile(@params);
 

LLama.Unittest/StatelessExecutorTest.cs

@@ -20,7 +20,7 @@ namespace LLama.Unittest
                 ContextSize = 60,
                 Seed = 1754,
                 BatchSize = 2,
-                GpuLayerCount = 0,                
+                GpuLayerCount = Constants.CIGpuLayerCount,                
             };
             _weights = LLamaWeights.LoadFromFile(_params);
         }

LLama.Unittest/TokenTests.cs

@@ -15,7 +15,7 @@ public sealed class TokenTests
         _params = new ModelParams(Constants.GenerativeModelPath)
         {
             ContextSize = 2048,
-            GpuLayerCount = 0,
+            GpuLayerCount = Constants.CIGpuLayerCount,
         };
         _model = LLamaWeights.LoadFromFile(_params);
     }