updated LLamaSharp to 0.7, updated examples

LangChain.Sources.slnf

@@ -5,7 +5,7 @@
       "src\\libs\\LangChain.Core\\LangChain.Core.csproj",
       "src\\libs\\Providers\\LangChain.Providers.LLamaSharp\\LangChain.Providers.LLamaSharp.csproj",
 	  "src\\libs\\Providers\\LangChain.Providers.HuggingFace\\LangChain.Providers.HuggingFace.csproj",
-	  "src\\tests\\LangChain.IntegrationTests.LLamaSharp\\LangChain.IntegrationTests.LLamaSharp.csproj",
+	  "src\\tests\\LangChain.Providers.LLamaSharp.IntegrationTests\\LangChain.Providers.LLamaSharp.IntegrationTests.csproj",
       "src\\tests\\LangChain.UnitTest\\LangChain.UnitTest.csproj"
     ]
   }

src/Directory.Packages.props

@@ -14,8 +14,8 @@
     <PackageVersion Include="H.Resources.Generator" Version="1.5.1" />
     <PackageVersion Include="HuggingFace" Version="0.2.4" />
     <PackageVersion Include="LeonardoAi" Version="0.1.0" />
-    <PackageVersion Include="LLamaSharp" Version="0.3.0" />
-    <PackageVersion Include="LLamaSharp.Backend.Cpu" Version="0.3.0" />
+    <PackageVersion Include="LLamaSharp" Version="0.7.0" />
+    <PackageVersion Include="LLamaSharp.Backend.Cpu" Version="0.7.0" />
     <PackageVersion Include="Microsoft.CodeAnalysis.CSharp.Workspaces" Version="4.7.0" />
     <PackageVersion Include="Microsoft.NET.Test.Sdk" Version="17.7.2" />
     <PackageVersion Include="Microsoft.SemanticKernel.Connectors.AI.OpenAI" Version="0.15.230531.5-preview" />

src/libs/Providers/LangChain.Providers.LLamaSharp/LLamaSharpConfiguration.cs

@@ -6,25 +6,22 @@ public class LLamaSharpConfiguration
     /// Path to *.bin file
     /// </summary>
     public string PathToModelFile { get; set; }
-
-    /// <summary>
-    /// Model mode.
-    /// Chat - for conversation completion
-    /// Instruction - for instruction execution
-    /// </summary>
-    public ELLamaSharpModelMode Mode { get; set; } = ELLamaSharpModelMode.Chat;
-
+
     /// <summary>
     /// Context size
     /// How much tokens model will remember.
     /// Usually 2048 for llama
     /// </summary>
-    public int ContextSize { get; set; } = 512;
+    public int ContextSize { get; set; } = 1024;
 
     /// <summary>
     /// Temperature
     /// The level of model's creativity
     /// </summary>
     public float Temperature { get; set; } = 0.7f;
 
+    public int Seed { get; set; }
+
+    public int MaxTokens { get; set; } = 600;
+
 }

src/libs/Providers/LangChain.Providers.LLamaSharp/LLamaSharpModelBase.cs

@@ -0,0 +1,50 @@
+using LLama.Common;
+using LLama;
+
+namespace LangChain.Providers.LLamaSharp;
+
+public abstract class LLamaSharpModelBase:IChatModel
+{
+    public string Id { get; }
+    public Usage TotalUsage { get; protected set; }
+    public int ContextLength =>_configuration.ContextSize;
+
+    protected readonly LLamaSharpConfiguration _configuration;
+    protected readonly LLamaWeights _model;
+    protected readonly ModelParams _parameters;
+
+    protected LLamaSharpModelBase(LLamaSharpConfiguration configuration)
+    {
+        _parameters = new ModelParams(configuration.PathToModelFile)
+        {
+            ContextSize = (uint)configuration.ContextSize,
+            Seed = (uint)configuration.Seed,
+
+        };
+        _model = LLamaWeights.LoadFromFile(_parameters);
+        _configuration = configuration;
+        Id = Path.GetFileNameWithoutExtension(configuration.PathToModelFile);
+    }
+
+    public abstract Task<ChatResponse>
+        GenerateAsync(ChatRequest request, CancellationToken cancellationToken = default);
+
+    protected string ConvertRole(MessageRole role)
+    {
+        return role switch
+        {
+            MessageRole.Human => "Human: ",
+            MessageRole.Ai => "Assistant: ",
+            MessageRole.System => "",
+            _ => throw new NotSupportedException($"the role {role} is not supported")
+        };
+    }
+    protected string ConvertMessage(Message message)
+    {
+        return $"{ConvertRole(message.Role)}{message.Content}";
+    }
+    protected string ToPrompt(IEnumerable<Message> messages)
+    {
+        return string.Join("\n", messages.Select(ConvertMessage).ToArray());
+    }
+}

src/libs/Providers/LangChain.Providers.LLamaSharp/LLamaSharpModelChat.cs

@@ -0,0 +1,65 @@
+using System.Diagnostics;
+using LLama;
+using System.Reflection;
+using LLama.Common;
+
+namespace LangChain.Providers.LLamaSharp
+{
+    public class LLamaSharpModelChat : LLamaSharpModelBase
+    {
+        public LLamaSharpModelChat(LLamaSharpConfiguration configuration) : base(configuration)
+        {
+        }
+
+
+        string SanitizeOutput(string res)
+        {
+            return res.Replace("Human:", "").Replace("Assistant:", "").Trim();
+        }
+
+        public override async Task<ChatResponse> GenerateAsync(ChatRequest request,
+            CancellationToken cancellationToken = default)
+        {
+            var prompt = ToPrompt(request.Messages);
+
+            var watch = Stopwatch.StartNew();
+
+
+            var context = _model.CreateContext(_parameters);
+            var ex = new InteractiveExecutor(context);
+            ChatSession session = new ChatSession(ex);
+            var inferenceParams = new InferenceParams()
+            {
+                Temperature = _configuration.Temperature,
+                AntiPrompts = new List<string> { "Human:" },
+                MaxTokens = _configuration.MaxTokens,
+
+            };
+
+
+            var buf = "";
+            await foreach (var text in session.ChatAsync(prompt,
+                               inferenceParams, cancellationToken))
+            {
+                buf += text;
+            }
+
+            buf = SanitizeOutput(buf);
+            var result = request.Messages.ToList();
+            result.Add(buf.AsAiMessage());
+
+            watch.Stop();
+
+            // Unsupported
+            var usage = Usage.Empty with
+            {
+                Time = watch.Elapsed,
+            };
+            TotalUsage += usage;
+
+            return new ChatResponse(
+                Messages: result,
+                Usage: usage);
+        }
+    }
+}

src/libs/Providers/LangChain.Providers.LLamaSharp/LLamaSharpModelInstruction.cs

@@ -0,0 +1,59 @@
+using LLama.Common;
+using LLama;
+using System.Diagnostics;
+
+namespace LangChain.Providers.LLamaSharp;
+
+public class LLamaSharpModelInstruction:LLamaSharpModelBase
+{
+    public LLamaSharpModelInstruction(LLamaSharpConfiguration configuration) : base(configuration)
+    {
+    }
+    string SanitizeOutput(string res)
+    {
+        return res.Replace("\n>", "").Trim();
+    }
+    public override async Task<ChatResponse> GenerateAsync(ChatRequest request, CancellationToken cancellationToken = default)
+    {
+        var prompt = ToPrompt(request.Messages)+"\n";
+
+        var watch = Stopwatch.StartNew();
+
+
+        var context = _model.CreateContext(_parameters);
+        var ex = new InstructExecutor(context);
+
+        var inferenceParams = new InferenceParams()
+        {
+            Temperature = _configuration.Temperature,
+            AntiPrompts = new List<string> { ">" },
+            MaxTokens = _configuration.MaxTokens,
+
+        };
+
+
+        var buf = "";
+        await foreach (var text in ex.InferAsync(prompt,
+                           inferenceParams, cancellationToken))
+        {
+            buf += text;
+        }
+
+        buf=SanitizeOutput(buf);
+        var result = request.Messages.ToList();
+        result.Add(buf.AsAiMessage());
+
+        watch.Stop();
+
+        // Unsupported
+        var usage = Usage.Empty with
+        {
+            Time = watch.Elapsed,
+        };
+        TotalUsage += usage;
+
+        return new ChatResponse(
+            Messages: result,
+            Usage: usage);
+    }
+}