Converted ModelParams into a `record` class. This has several advantages:

- Equality, hashing etc all implemented automatically
 - Default values are defined in just one place (the properties) instead of the constructor as well
 - Added test to ensure that serialization works properly

LLama.Examples/NewVersion/ChatSessionStripRoleName.cs

@@ -10,7 +10,12 @@ namespace LLama.Examples.NewVersion
             var modelPath = Console.ReadLine();
             var prompt = File.ReadAllText("Assets/chat-with-bob.txt").Trim();
 
-            var parameters = new ModelParams(modelPath, contextSize: 1024, seed: 1337, gpuLayerCount: 5);
+            var parameters = new ModelParams(modelPath)
+            {
+                ContextSize = 1024,
+                Seed = 1337,
+                GpuLayerCount = 5
+            };
             using var model = LLamaWeights.LoadFromFile(parameters);
             using var context = model.CreateContext(parameters);
             var executor = new InteractiveExecutor(context);

LLama.Examples/NewVersion/ChatSessionWithRoleName.cs

@@ -10,7 +10,12 @@ namespace LLama.Examples.NewVersion
             var modelPath = Console.ReadLine();
             var prompt = File.ReadAllText("Assets/chat-with-bob.txt").Trim();
 
-            var parameters = new ModelParams(modelPath, contextSize: 1024, seed: 1337, gpuLayerCount: 5);
+            var parameters = new ModelParams(modelPath)
+            {
+                ContextSize = 1024,
+                Seed = 1337,
+                GpuLayerCount = 5
+            };
             using var model = LLamaWeights.LoadFromFile(parameters);
             using var context = model.CreateContext(parameters);
             var executor = new InteractiveExecutor(context);

LLama.Examples/NewVersion/InstructModeExecute.cs

@@ -10,7 +10,12 @@ namespace LLama.Examples.NewVersion
             var modelPath = Console.ReadLine();
             var prompt = File.ReadAllText("Assets/dan.txt").Trim();
 
-            var parameters = new ModelParams(modelPath, contextSize: 1024, seed: 1337, gpuLayerCount: 5);
+            var parameters = new ModelParams(modelPath)
+            {
+                ContextSize = 1024,
+                Seed = 1337,
+                GpuLayerCount = 5
+            };
             using var model = LLamaWeights.LoadFromFile(parameters);
             using var context = model.CreateContext(parameters);
             var executor = new InstructExecutor(context);

LLama.Examples/NewVersion/InteractiveModeExecute.cs

@@ -10,7 +10,12 @@ namespace LLama.Examples.NewVersion
             var modelPath = Console.ReadLine();
             var prompt = (await File.ReadAllTextAsync("Assets/chat-with-bob.txt")).Trim();
 
-            var parameters = new ModelParams(modelPath, contextSize: 1024, seed: 1337, gpuLayerCount: 5);
+            var parameters = new ModelParams(modelPath)
+            {
+                ContextSize = 1024,
+                Seed = 1337,
+                GpuLayerCount = 5
+            };
             using var model = LLamaWeights.LoadFromFile(parameters);
             using var context = model.CreateContext(parameters);
             var ex = new InteractiveExecutor(context);

LLama.Examples/NewVersion/LoadAndSaveSession.cs

@@ -10,7 +10,12 @@ namespace LLama.Examples.NewVersion
             var modelPath = Console.ReadLine();
             var prompt = File.ReadAllText("Assets/chat-with-bob.txt").Trim();
 
-            var parameters = new ModelParams(modelPath, contextSize: 1024, seed: 1337, gpuLayerCount: 5);
+            var parameters = new ModelParams(modelPath)
+            {
+                ContextSize = 1024,
+                Seed = 1337,
+                GpuLayerCount = 5
+            };
             using var model = LLamaWeights.LoadFromFile(parameters);
             using var context = model.CreateContext(parameters);
             var ex = new InteractiveExecutor(context);
@@ -45,7 +50,7 @@ namespace LLama.Examples.NewVersion
                     Console.ForegroundColor = ConsoleColor.White;
 
                     ex.Context.Dispose();
-                    ex = new(new LLamaContext(new ModelParams(modelPath, contextSize: 1024, seed: 1337, gpuLayerCount: 5)));
+                    ex = new(new LLamaContext(parameters));
                     session = new ChatSession(ex);
                     session.LoadSession(statePath);
 

LLama.Examples/NewVersion/LoadAndSaveState.cs

@@ -10,7 +10,12 @@ namespace LLama.Examples.NewVersion
             var modelPath = Console.ReadLine();
             var prompt = File.ReadAllText("Assets/chat-with-bob.txt").Trim();
 
-            var parameters = new ModelParams(modelPath, contextSize: 1024, seed: 1337, gpuLayerCount: 5);
+            var parameters = new ModelParams(modelPath)
+            {
+                ContextSize = 1024,
+                Seed = 1337,
+                GpuLayerCount = 5
+            };
             using var model = LLamaWeights.LoadFromFile(parameters);
             using var context = model.CreateContext(parameters);
             var ex = new InteractiveExecutor(context);

LLama.Examples/NewVersion/StatelessModeExecute.cs

@@ -9,7 +9,12 @@ namespace LLama.Examples.NewVersion
             Console.Write("Please input your model path: ");
             var modelPath = Console.ReadLine();
 
-            var parameters = new ModelParams(modelPath, contextSize: 1024, seed: 1337, gpuLayerCount: 5);
+            var parameters = new ModelParams(modelPath)
+            {
+                ContextSize = 1024,
+                Seed = 1337,
+                GpuLayerCount = 5
+            };
             using var model = LLamaWeights.LoadFromFile(parameters);
             var ex = new StatelessExecutor(model, parameters);
 

LLama.Unittest/BasicTest.cs

@@ -10,7 +10,10 @@ namespace LLama.Unittest
 
         public BasicTest()
         {
-            _params = new ModelParams("Models/llama-2-7b-chat.ggmlv3.q3_K_S.bin", contextSize: 2048);
+            _params = new ModelParams("Models/llama-2-7b-chat.ggmlv3.q3_K_S.bin")
+            {
+                ContextSize = 2048
+            };
             _model = LLamaWeights.LoadFromFile(_params);
         }
 

LLama.Unittest/GrammarTest.cs

@@ -11,7 +11,10 @@ namespace LLama.Unittest
 
         public GrammarTest()
         {
-            _params = new ModelParams("Models/llama-2-7b-chat.ggmlv3.q3_K_S.bin", contextSize: 2048);
+            _params = new ModelParams("Models/llama-2-7b-chat.ggmlv3.q3_K_S.bin")
+            {
+                ContextSize = 2048,
+            };
             _model = LLamaWeights.LoadFromFile(_params);
         }
 

LLama.Unittest/ModelsParamsTests.cs

@@ -0,0 +1,45 @@
+using LLama.Common;
+
+namespace LLama.Unittest
+{
+    public class ModelsParamsTests
+    {
+        [Fact]
+        public void SerializeRoundTripSystemTextJson()
+        {
+            var expected = new ModelParams("abc/123")
+            {
+                BatchSize = 17,
+                ContextSize = 42,
+                LoraAdapter = "adapter",
+                GroupedQueryAttention = 7,
+                Seed = 42,
+                GpuLayerCount = 111
+            };
+
+            var json = System.Text.Json.JsonSerializer.Serialize(expected);
+            var actual = System.Text.Json.JsonSerializer.Deserialize<ModelParams>(json);
+
+            Assert.Equal(expected, actual);
+        }
+
+        [Fact]
+        public void SerializeRoundTripNewtonsoft()
+        {
+            var expected = new ModelParams("abc/123")
+            {
+                BatchSize = 17,
+                ContextSize = 42,
+                LoraAdapter = "adapter",
+                GroupedQueryAttention = 7,
+                Seed = 42,
+                GpuLayerCount = 111
+            };
+
+            var json = Newtonsoft.Json.JsonConvert.SerializeObject(expected);
+            var actual = Newtonsoft.Json.JsonConvert.DeserializeObject<ModelParams>(json);
+
+            Assert.Equal(expected, actual);
+        }
+    }
+}

LLama.WebAPI/Services/StatefulChatService.cs

@@ -16,7 +16,10 @@ public class StatefulChatService : IDisposable
 
     public StatefulChatService(IConfiguration configuration)
     {
-        _context = new LLamaContext(new Common.ModelParams(configuration["ModelPath"], contextSize: 512));
+        _context = new LLamaContext(new Common.ModelParams(configuration["ModelPath"])
+        {
+            ContextSize = 512
+        });
         _session = new ChatSession(new InteractiveExecutor(_context));
     }
 

LLama.WebAPI/Services/StatelessChatService.cs

@@ -12,7 +12,10 @@ namespace LLama.WebAPI.Services
 
         public StatelessChatService(IConfiguration configuration)
         {
-            _context = new LLamaContext(new ModelParams(configuration["ModelPath"], contextSize: 512));
+            _context = new LLamaContext(new ModelParams(configuration["ModelPath"])
+            {
+                ContextSize = 512,
+            });
             // TODO: replace with a stateless executor
             _session = new ChatSession(new InteractiveExecutor(_context))
                         .WithOutputTransform(new LLamaTransforms.KeywordTextOutputStreamTransform(new string[] { "User:", "Assistant:" }, redundancyLength: 8))

LLama/Common/ModelParams.cs

@@ -6,7 +6,7 @@ namespace LLama.Common
     /// <summary>
     /// The parameters for initializing a LLama model.
     /// </summary>
-    public class ModelParams
+    public record ModelParams
         : IModelParams
     {
         /// <summary>
@@ -116,6 +116,22 @@ namespace LLama.Common
         /// </summary>
         public string Encoding { get; set; } = "UTF-8";
 
+        /// <summary>
+        /// 
+        /// </summary>
+        /// <param name="modelPath">The model path.</param>
+        [System.Text.Json.Serialization.JsonConstructor]
+        public ModelParams(string modelPath)
+        {
+            ModelPath = modelPath;
+        }
+
+        private ModelParams()
+        {
+            // This constructor (default parameterless constructor) is used by Newtonsoft to deserialize!
+            ModelPath = "";
+        }
+
         /// <summary>
         /// 
         /// </summary>
@@ -139,6 +155,7 @@ namespace LLama.Common
         /// <param name="ropeFrequencyScale">RoPE frequency scaling factor</param>
         /// <param name="mulMatQ">Use experimental mul_mat_q kernels</param>
         /// <param name="encoding">The encoding to use to convert text for the model</param>
+        [Obsolete("Use object initializer to set all optional parameters")]
         public ModelParams(string modelPath, int contextSize = 512, int gpuLayerCount = 20,
                            int seed = 1337, bool useFp16Memory = true,
                            bool useMemorymap = true, bool useMemoryLock = false, bool perplexity = false,
@@ -161,10 +178,10 @@ namespace LLama.Common
             BatchSize = batchSize;
             ConvertEosToNewLine = convertEosToNewLine;
             EmbeddingMode = embeddingMode;
-            GroupedQueryAttention  = groupedQueryAttention;
+            GroupedQueryAttention = groupedQueryAttention;
             RmsNormEpsilon = rmsNormEpsilon;
-            RopeFrequencyBase  = ropeFrequencyBase;
-            RopeFrequencyScale  = ropeFrequencyScale;
+            RopeFrequencyBase = ropeFrequencyBase;
+            RopeFrequencyScale = ropeFrequencyScale;
             MulMatQ = mulMatQ;
             Encoding = encoding;
         }