Merge pull request #606 from iceljc/features/add-embedding-dim-setting

Features/add embedding dim setting

Author: iceljc

src/Infrastructure/BotSharp.Abstraction/Knowledges/IKnowledgeService.cs

@@ -8,8 +8,10 @@ public interface IKnowledgeService
     Task<IEnumerable<string>> GetVectorCollections();
     Task<IEnumerable<VectorSearchResult>> SearchVectorKnowledge(string query, string collectionName, VectorSearchOptions options);
     Task FeedVectorKnowledge(string collectionName, KnowledgeCreationModel model);
-    Task<StringIdPagedItems<VectorSearchResult>> GetVectorCollectionData(string collectionName, VectorFilter filter);
+    Task<StringIdPagedItems<VectorSearchResult>> GetPagedVectorCollectionData(string collectionName, VectorFilter filter);
     Task<bool> DeleteVectorCollectionData(string collectionName, string id);
+    Task<bool> CreateVectorCollectionData(string collectionName, VectorCreateModel create);
+    Task<bool> UpdateVectorCollectionData(string collectionName, VectorUpdateModel update);
     Task<GraphSearchResult> SearchGraphKnowledge(string query, GraphSearchOptions options);
     Task<KnowledgeSearchResult> SearchKnowledge(string query, string collectionName, VectorSearchOptions vectorOptions, GraphSearchOptions graphOptions);
 }

src/Infrastructure/BotSharp.Abstraction/Knowledges/Settings/KnowledgeBaseSettings.cs

@@ -4,14 +4,28 @@ namespace BotSharp.Abstraction.Knowledges.Settings;
 
 public class KnowledgeBaseSettings
 {
-    public string DefaultCollection { get; set; } = KnowledgeCollectionName.BotSharp;
     public string VectorDb { get; set; }
     public string GraphDb { get; set; }
-    public KnowledgeModelSetting TextEmbedding { get; set; }
+
+    public DefaultKnowledgeBaseSetting Default { get; set; }
+    public List<VectorCollectionSetting> Collections { get; set; } = new();
+}
+
+public class DefaultKnowledgeBaseSetting
+{
+    public string CollectionName { get; set; } = KnowledgeCollectionName.BotSharp;
+    public KnowledgeTextEmbeddingSetting TextEmbedding { get; set; }
+}
+
+public class VectorCollectionSetting
+{
+    public string Name { get; set; }
+    public KnowledgeTextEmbeddingSetting TextEmbedding { get; set; }
 }
 
-public class KnowledgeModelSetting
+public class KnowledgeTextEmbeddingSetting
 {
     public string Provider { get; set; }
     public string Model { get; set; }
+    public int Dimension { get; set; }
 }

src/Infrastructure/BotSharp.Abstraction/MLTasks/ITextEmbedding.cs

@@ -6,8 +6,9 @@ public interface ITextEmbedding
     /// The Embedding provider like Microsoft Azure, OpenAI, ClaudAI
     /// </summary>
     string Provider { get; }
-    int Dimension { get; set; }
     Task<float[]> GetVectorAsync(string text);
     Task<List<float[]>> GetVectorsAsync(List<string> texts);
     void SetModelName(string model);
+    void SetDimension(int dimension);
+    int GetDimension();
 }

src/Infrastructure/BotSharp.Abstraction/MLTasks/Settings/LlmModelSetting.cs

@@ -52,6 +52,11 @@ public class LlmModelSetting
     /// </summary>
     public float CompletionCost { get; set; }
 
+    /// <summary>
+    /// Embedding dimension
+    /// </summary>
+    public int Dimension { get; set; }
+
     public override string ToString()
     {
         return $"[{Type}] {Name} {Endpoint}";

src/Infrastructure/BotSharp.Abstraction/VectorStorage/IVectorDb.cs

@@ -7,7 +7,8 @@ public interface IVectorDb
     string Name { get; }
 
     Task<IEnumerable<string>> GetCollections();
-    Task<StringIdPagedItems<VectorCollectionData>> GetCollectionData(string collectionName, VectorFilter filter);
+    Task<StringIdPagedItems<VectorCollectionData>> GetPagedCollectionData(string collectionName, VectorFilter filter);
+    Task<IEnumerable<VectorCollectionData>> GetCollectionData(string collectionName, IEnumerable<Guid> ids, bool withPayload = false, bool withVector = false);
     Task CreateCollection(string collectionName, int dim);
     Task<bool> Upsert(string collectionName, Guid id, float[] vector, string text, Dictionary<string, string>? payload = null);
     Task<IEnumerable<VectorCollectionData>> Search(string collectionName, float[] vector, IEnumerable<string>? fields, int limit = 5, float confidence = 0.5f, bool withVector = false);

src/Infrastructure/BotSharp.Abstraction/VectorStorage/Models/VectorCreateModel.cs

@@ -0,0 +1,7 @@
+namespace BotSharp.Abstraction.VectorStorage.Models;
+
+public class VectorCreateModel
+{
+    public string Text { get; set; }
+    public Dictionary<string, string>? Payload { get; set; }
+}

src/Infrastructure/BotSharp.Abstraction/VectorStorage/Models/VectorUpdateModel.cs

@@ -0,0 +1,6 @@
+namespace BotSharp.Abstraction.VectorStorage.Models;
+
+public class VectorUpdateModel : VectorCreateModel
+{
+    public string Id { get; set; }
+}

src/Infrastructure/BotSharp.Core/Infrastructures/CompletionProvider.cs

@@ -111,7 +111,12 @@ public static ITextEmbedding GetTextEmbedding(IServiceProvider services,
             logger.LogError($"Can't resolve completion provider by {provider}");
         }
 
+
+        var llmProviderService = services.GetRequiredService<ILlmProviderService>();
+        var found = llmProviderService.GetSetting(provider, model);
+
         completer.SetModelName(model);
+        completer.SetDimension(found.Dimension);
         return completer;
     }
 

src/Infrastructure/BotSharp.OpenAPI/Controllers/KnowledgeBaseController.cs

@@ -39,10 +39,10 @@ public async Task<IEnumerable<VectorKnowledgeViewModel>> SearchVectorKnowledge([
         return results.Select(x => VectorKnowledgeViewModel.From(x)).ToList();
     }
 
-    [HttpPost("/knowledge/vector/{collection}/data")]
-    public async Task<StringIdPagedItems<VectorKnowledgeViewModel>> GetVectorCollectionData([FromRoute] string collection, [FromBody] VectorFilter filter)
+    [HttpPost("/knowledge/vector/{collection}/page")]
+    public async Task<StringIdPagedItems<VectorKnowledgeViewModel>> GetPagedVectorCollectionData([FromRoute] string collection, [FromBody] VectorFilter filter)
     {
-        var data = await _knowledgeService.GetVectorCollectionData(collection, filter);
+        var data = await _knowledgeService.GetPagedVectorCollectionData(collection, filter);
         var items = data.Items?.Select(x => VectorKnowledgeViewModel.From(x))?
                                .ToList() ?? new List<VectorKnowledgeViewModel>();
 
@@ -54,6 +54,33 @@ public async Task<StringIdPagedItems<VectorKnowledgeViewModel>> GetVectorCollect
         };
     }
 
+    [HttpPost("/knowledge/vector/{collection}/create")]
+    public async Task<bool> CreateVectorKnowledge([FromRoute] string collection, [FromBody] VectorKnowledgeCreateRequest request)
+    {
+        var create = new VectorCreateModel
+        {
+            Text = request.Text,
+            Payload = request.Payload
+        };
+
+        var created = await _knowledgeService.CreateVectorCollectionData(collection, create);
+        return created;
+    }
+
+    [HttpPut("/knowledge/vector/{collection}/update")]
+    public async Task<bool> UpdateVectorKnowledge([FromRoute] string collection, [FromBody] VectorKnowledgeUpdateRequest request)
+    {
+        var update = new VectorUpdateModel
+        {
+            Id = request.Id,
+            Text = request.Text,
+            Payload = request.Payload
+        };
+
+        var updated = await _knowledgeService.UpdateVectorCollectionData(collection, update);
+        return updated;
+    }
+
     [HttpDelete("/knowledge/vector/{collection}/data/{id}")]
     public async Task<bool> DeleteVectorCollectionData([FromRoute] string collection, [FromRoute] string id)
     {

src/Infrastructure/BotSharp.OpenAPI/Controllers/TextEmbeddingController.cs

@@ -18,7 +18,7 @@ public TextEmbeddingController(
         _logger = logger;
     }
 
-    [HttpPost("/text-embedding/generation")]
+    [HttpPost("/text-embedding/generate")]
     public async Task<List<float[]>> GenerateTextEmbeddings(EmbeddingInputModel input)
     {
         var state = _services.GetRequiredService<IConversationStateService>();
@@ -27,7 +27,10 @@ public async Task<List<float[]>> GenerateTextEmbeddings(EmbeddingInputModel inpu
         try
         {
             var completion = CompletionProvider.GetTextEmbedding(_services, provider: input.Provider ?? "openai", model: input.Model ?? "text-embedding-3-large");
-            completion.Dimension = input.Dimension;
+            if (input.Dimension.HasValue && input.Dimension.Value > 0)
+            {
+                completion.SetDimension(input.Dimension.Value);
+            }
 
             var embeddings = await completion.GetVectorsAsync(input.Texts?.ToList() ?? []);
             return embeddings;