Merge pull request #52 from wisedev-code/feat/migrate-to-new-release-and-km-imprv

wisedev-pstach · web-flow · commit bbf23e60d96f · 2025-05-16T15:30:09.000+02:00
Feat/migrate to new release and km imprv
diff --git a/Examples/Examples/Agents/AgentWithBecomeExample.cs b/Examples/Examples/Agents/AgentWithBecomeExample.cs
@@ -1,5 +1,6 @@
 using MaIN.Core.Hub;
 using MaIN.Core.Hub.Utils;
+using MaIN.Domain.Entities;
 using MaIN.Domain.Entities.Agents.AgentSource;
 using MaIN.Domain.Models;
 
@@ -17,17 +18,17 @@ public async Task Start()
                 Path = "./Files/Books.json",
                 Name = "Books.json"
             }, AgentSourceType.File)
-            .WithBehaviour("SalesGod", 
+            .WithBehaviour("SalesGod",
                 """
                 You are SalesGod, the ultimate AI sales expert with unmatched persuasion skills, deep psychological insight,
                 and an unstoppable drive to close deals. Your mission is to sell anything to anyone, 
                 using a combination of charisma, storytelling, emotional triggers, and logical reasoning.
                 Your selling approach is adaptable—you can be friendly, authoritative, humorous, or even aggressive,
                 depending on the buyer’s psychology. You master every sales technique, from scarcity and urgency to social proof and objection handling.
-                
+
                 No hesitation. No doubts. Every conversation is an opportunity to seal the deal. You never give up,
                 always finding a way to turn ‘no’ into ‘yes.’ Now, go out there and SELL!
-                
+
                 Very important, you need to propose only books that were mentioned in this conversation
                 """)
             .WithSteps(StepBuilder.Instance
@@ -36,9 +37,8 @@ public async Task Start()
                 .Answer()
                 .Build())
             .Create(interactiveResponse: true);
-        
+
         await becomeAgent
             .ProcessAsync("I am looking for good fantasy book to buy");
-
     }
 }
diff --git a/Releases/0.2.2.md b/Releases/0.2.2.md
@@ -0,0 +1,6 @@
+# 0.2.2 release
+
+- Added Gemini support in infer cli command
+- Package upgrades
+- New CLI release (support for Qwen3-8b, Qwen3-14 and Phi3.5 and Phi4)
+- Multimodal and KM improvements
diff --git a/src/MaIN.Core/.nuspec b/src/MaIN.Core/.nuspec
@@ -2,7 +2,7 @@
 <package>
   <metadata>
     <id>MaIN.NET</id>
-    <version>0.2.1</version>
+    <version>0.2.2</version>
     <authors>Wisedev</authors>
     <owners>Wisedev</owners>
     <icon>favicon.png</icon>
@@ -13,9 +13,9 @@
     <dependencies>
       <dependency id="GTranslate" version="2.2.0" />
       <dependency id="HtmlAgilityPack" version="1.12.0" />
-      <dependency id="LLamaSharp" version="0.23.0" />
-      <dependency id="LLamaSharp.Backend.Cuda12" version="0.23.0" />
-      <dependency id="LLamaSharp.kernel-memory" version="0.23.0" />
+      <dependency id="LLamaSharp" version="0.24.0" />
+      <dependency id="LLamaSharp.Backend.Cuda12" version="0.24.0" />
+      <dependency id="LLamaSharp.kernel-memory" version="0.24.0" />
       <dependency id="Microsoft.Extensions.Http" version="9.0.3" />
       <dependency id="Microsoft.KernelMemory" version="0.98.250324.1" />
       <dependency id="System.Data.SqlClient" version="4.8.6" />
diff --git a/src/MaIN.Core/MaIN.Core.csproj b/src/MaIN.Core/MaIN.Core.csproj
@@ -9,9 +9,9 @@
     <ItemGroup>
         <PackageReference Include="GTranslate" Version="2.2.0" />
         <PackageReference Include="HtmlAgilityPack" Version="1.12.1" />
-        <PackageReference Include="LLamaSharp" Version="0.23.0" />
-        <PackageReference Include="LLamaSharp.Backend.Cuda12" Version="0.23.0" />
-        <PackageReference Include="LLamaSharp.kernel-memory" Version="0.23.0" />
+        <PackageReference Include="LLamaSharp" Version="0.24.0" />
+        <PackageReference Include="LLamaSharp.Backend.Cuda12" Version="0.24.0" />
+        <PackageReference Include="LLamaSharp.kernel-memory" Version="0.24.0" />
         <PackageReference Include="Microsoft.Extensions.Http" Version="9.0.3" />
         <PackageReference Include="Microsoft.KernelMemory" Version="0.98.250324.1" />
         <PackageReference Include="System.Data.SqlClient" Version="4.8.6" />
diff --git a/src/MaIN.Domain/Entities/MemoryParams.cs b/src/MaIN.Domain/Entities/MemoryParams.cs
@@ -14,4 +14,6 @@ public class MemoryParams
     /// If your response is invalid make sure you meet those limits.
     /// </summary>
     public int AnswerTokens { get; set; } = 500;
+
+    public bool MultiModalMode { get; set; } = false;
 }
diff --git a/src/MaIN.Domain/MaIN.Domain.csproj b/src/MaIN.Domain/MaIN.Domain.csproj
@@ -8,7 +8,7 @@
     </PropertyGroup>
 
     <ItemGroup>
-      <PackageReference Include="LLamaSharp" Version="0.23.0" />
+      <PackageReference Include="LLamaSharp" Version="0.24.0" />
     </ItemGroup>
 
 </Project>
diff --git a/src/MaIN.Domain/Models/SupportedModels.cs b/src/MaIN.Domain/Models/SupportedModels.cs
@@ -38,6 +38,22 @@ public static class KnownModels
             DownloadUrl = string.Empty,
         },
         new Model()
+        {
+            Description = string.Empty,
+            Name = KnownModelNames.Qwen3_8b,
+            FileName = "Qwen3-8b.gguf",
+            DownloadUrl = string.Empty,
+            ReasonFunction = ReasoningFunctions.ProcessDeepSeekToken
+        },
+        new Model()
+        {
+            Description = string.Empty,
+            Name = KnownModelNames.Qwen3_14b,
+            FileName = "Qwen3-14b.gguf",
+            DownloadUrl = string.Empty,
+            ReasonFunction = ReasoningFunctions.ProcessDeepSeekToken
+        },
+        new Model()
         {
             Description = string.Empty,
             Name = KnownModelNames.OlympicCoder_7b,
@@ -108,6 +124,20 @@ public static class KnownModels
             DownloadUrl = string.Empty
         },
         new Model()
+        {
+            Description = string.Empty,
+            Name = KnownModelNames.Phi3_5_3b,
+            FileName = "phi3.5-3b.gguf",
+            DownloadUrl = string.Empty
+        },
+        new Model()
+        {
+            Description = string.Empty,
+            Name = KnownModelNames.Phi4_4b,
+            FileName = "phi4-4b.gguf",
+            DownloadUrl = string.Empty
+        },
+        new Model()
         {
             Description = string.Empty,
             Name = KnownModelNames.DeepSeek_R1_8b,
@@ -247,7 +277,11 @@ public struct KnownModelNames
     public const string DeepSeek_R1_8b = "deepseekR1:8b";
     public const string DeepSeek_R1_1_5b = "deepseekR1:1.5b";
     public const string QwQ_7b = "qwq:7b";
+    public const string Qwen3_8b = "qwen3:8b";
+    public const string Qwen3_14b = "qwen3:14b";
     public const string Olmo2_7b = "olmo2:7b";
+    public const string Phi3_5_3b = "phi3.5:3b";
+    public const string Phi4_4b = "phi4:4b";
     public const string Smollm2_0_1b = "smollm2:0.1b";
     public const string Yi_6b = "yi:6b";
 
diff --git a/src/MaIN.InferPage/Program.cs b/src/MaIN.InferPage/Program.cs
@@ -81,6 +81,13 @@
         settings.BackendType = BackendType.OpenAi;
     });
 }
+if (Utils.Gemini)
+{
+    builder.Services.AddMaIN(builder.Configuration, settings =>
+    {
+        settings.BackendType = BackendType.Gemini;
+    });
+}
 else
 {
     builder.Services.AddMaIN(builder.Configuration);
diff --git a/src/MaIN.Infrastructure/Models/MemoryParamsDocument.cs b/src/MaIN.Infrastructure/Models/MemoryParamsDocument.cs
@@ -8,4 +8,5 @@ public class MemoryParamsDocument
     public float FrequencyPenalty { get; set; }
     public float Temperature { get; set; } 
     public int AnswerTokens { get; set; }
+    public bool MultiModalMode { get; set; }
 }
diff --git a/src/MaIN.Services/MaIN.Services.csproj b/src/MaIN.Services/MaIN.Services.csproj
@@ -15,9 +15,9 @@
     <ItemGroup>
       <PackageReference Include="GTranslate" Version="2.2.0" />
       <PackageReference Include="HtmlAgilityPack" Version="1.12.1" />
-      <PackageReference Include="LLamaSharp" Version="0.23.0" />
-      <PackageReference Include="LLamaSharp.Backend.Cuda12" Version="0.23.0" />
-      <PackageReference Include="LLamaSharp.kernel-memory" Version="0.23.0" />
+      <PackageReference Include="LLamaSharp" Version="0.24.0" />
+      <PackageReference Include="LLamaSharp.Backend.Cuda12" Version="0.24.0" />
+      <PackageReference Include="LLamaSharp.kernel-memory" Version="0.24.0" />
       <PackageReference Include="Microsoft.Extensions.Http" Version="9.0.3" />
       <PackageReference Include="Microsoft.KernelMemory" Version="0.98.250324.1" />
       <PackageReference Include="Microsoft.SemanticKernel" Version="1.49.0" />
diff --git a/src/MaIN.Services/Mappers/ChatMapper.cs b/src/MaIN.Services/Mappers/ChatMapper.cs
@@ -90,6 +90,7 @@ public static ChatDocument ToDocument(this Chat chat)
             Messages = chat.Messages.Select(m => m.ToDocument()).ToList(),
             Visual = chat.Visual,
             Backend = chat.Backend,
+            MemoryParams = chat.MemoryParams.ToDocument(),
             InferenceParams = chat.InterferenceParams.ToDocument(),
             ConvState = chat.ConversationState,
             Properties = chat.Properties,
@@ -109,6 +110,7 @@ public static Chat ToDomain(this ChatDocument chat)
             Backend = chat.Backend,
             Properties = chat.Properties,
             ConversationState = chat.ConvState as Conversation.State,
+            MemoryParams = chat.MemoryParams!.ToDomain(),
             InterferenceParams = chat.InferenceParams!.ToDomain(),
             Interactive = chat.Interactive,
             Translate = chat.Translate,
@@ -164,6 +166,7 @@ private static MemoryParams ToDomain(this MemoryParamsDocument memoryParams)
         {
             Temperature = memoryParams.Temperature,
             AnswerTokens = memoryParams.AnswerTokens,
+            MultiModalMode = memoryParams.MultiModalMode,
             ContextSize = memoryParams.ContextSize,
             GpuLayerCount = memoryParams.GpuLayerCount,
             MaxMatchesCount = memoryParams.MaxMatchesCount,
@@ -193,6 +196,7 @@ private static MemoryParamsDocument ToDocument(this MemoryParams memoryParams)
         {
             Temperature = memoryParams.Temperature,
             AnswerTokens = memoryParams.AnswerTokens,
+            MultiModalMode = memoryParams.MultiModalMode,
             ContextSize = memoryParams.ContextSize,
             GpuLayerCount = memoryParams.GpuLayerCount,
             MaxMatchesCount = memoryParams.MaxMatchesCount,
diff --git a/src/MaIN.Services/Services/LLMService/LLMService.cs b/src/MaIN.Services/Services/LLMService/LLMService.cs
@@ -99,13 +99,14 @@ public Task CleanSessionCache(string? id)
             Embeddings = true
         };
         var disableCache = chat.Properties.CheckProperty(ServiceConstants.Properties.DisableCacheProperty);
-        using var llmModel = disableCache
+        var llmModel = disableCache
             ? await LLamaWeights.LoadFromFileAsync(parameters, cancellationToken)
             : await ModelLoader.GetOrLoadModelAsync(modelsPath, model.FileName);
 
         var memory = memoryFactory.CreateMemoryWithModel(
             modelsPath,
             llmModel,
+            model.FileName,
             chat.MemoryParams);
 
         await memoryService.ImportDataToMemory(memory.KM, memoryOptions, cancellationToken);
@@ -114,7 +115,15 @@ public Task CleanSessionCache(string? id)
             userMessage.Content,
             cancellationToken: cancellationToken);
         await memory.KM.DeleteIndexAsync(cancellationToken: cancellationToken);
+
+        if (disableCache)
+        {
+            llmModel.Dispose();
+        }
+        
         memory.TextGenerationContext.Dispose();
+        memory.EmbeddingGenerator.Dispose();
+
         return new ChatResult
         {
             Done = true,
diff --git a/src/MaIN.Services/Services/LLMService/Memory/IMemoryFactory.cs b/src/MaIN.Services/Services/LLMService/Memory/IMemoryFactory.cs
@@ -1,14 +1,15 @@
 using LLama;
+using LLamaSharp.KernelMemory;
 using MaIN.Domain.Entities;
 using Microsoft.KernelMemory;
 
 namespace MaIN.Services.Services.LLMService.Memory;
 
 public interface IMemoryFactory
 {
-    IKernelMemory CreateMemory(string modelsPath, string modelName);
-    (IKernelMemory KM, LLamaContext TextGenerationContext) CreateMemoryWithModel(string modelsPath, LLamaWeights llmModel,
-        MemoryParams memoryParams);
+    (IKernelMemory KM, LLamaContext TextGenerationContext, LLamaSharpTextEmbeddingGenerator EmbeddingGenerator)
+        CreateMemoryWithModel(string modelsPath, LLamaWeights llmModel, string modelName,
+            MemoryParams memoryParams);
     IKernelMemory CreateMemoryWithOpenAi(string openAiKey, MemoryParams memoryParams);
     IKernelMemory CreateMemoryWithGemini(string geminiKey, MemoryParams memoryParams);
 }
diff --git a/src/MaIN.Services/Services/LLMService/Memory/MemoryFactory.cs b/src/MaIN.Services/Services/LLMService/Memory/MemoryFactory.cs
diff --git a/src/MaIN/MaIN.csproj b/src/MaIN/MaIN.csproj

Original file line number	Diff line number	Diff line change
`@@ -14,4 +14,6 @@ public class MemoryParams`
`14`	`14`	`/// If your response is invalid make sure you meet those limits.`
`15`	`15`	`/// </summary>`
`16`	`16`	`public int AnswerTokens { get; set; } = 500;`
	`17`	`+`
	`18`	`+ public bool MultiModalMode { get; set; } = false;`
`17`	`19`	`}`
Original file line number	Diff line number	Diff line change
`@@ -81,6 +81,13 @@`
`81`	`81`	`settings.BackendType = BackendType.OpenAi;`
`82`	`82`	`});`
`83`	`83`	`}`
	`84`	`+if (Utils.Gemini)`
	`85`	`+{`
	`86`	`+ builder.Services.AddMaIN(builder.Configuration, settings =>`
	`87`	`+ {`
	`88`	`+ settings.BackendType = BackendType.Gemini;`
	`89`	`+ });`
	`90`	`+}`
`84`	`91`	`else`
`85`	`92`	`{`
`86`	`93`	`builder.Services.AddMaIN(builder.Configuration);`
Original file line number	Diff line number	Diff line change
`@@ -8,4 +8,5 @@ public class MemoryParamsDocument`
`8`	`8`	`public float FrequencyPenalty { get; set; }`
`9`	`9`	`public float Temperature { get; set; }`
`10`	`10`	`public int AnswerTokens { get; set; }`
	`11`	`+ public bool MultiModalMode { get; set; }`
`11`	`12`	`}`