Fix: Load embedding model when loading existing Qdrant collections

- Qdrant get_collection now retrieves embedding model from document metadata
- Store embedding_model and chunking_strategy in Qdrant document payloads
- Extract embedding model to session state for both ChromaDB and Qdrant
- Switch theme to light mode

.streamlit/config.toml

@@ -20,5 +20,5 @@ serverPort = 7860
 level = "warning"
 
 [theme]
-base = "dark"
+base = "light"
 primaryColor = "#7C3AED"

streamlit_app.py

@@ -557,6 +557,27 @@ def load_existing_collection(api_key: str, collection_name: str, llm_provider: s
             st.session_state.collection_name = collection_name
             st.session_state.llm_provider = llm_provider
             
+            # Extract embedding model from collection metadata or vector store
+            embedding_model_name = None
+            
+            # For ChromaDB: check collection metadata
+            if hasattr(vector_store, 'current_collection') and vector_store.current_collection:
+                if hasattr(vector_store.current_collection, 'metadata'):
+                    collection_metadata = vector_store.current_collection.metadata
+                    if collection_metadata and "embedding_model" in collection_metadata:
+                        embedding_model_name = collection_metadata["embedding_model"]
+            
+            # For Qdrant or fallback: check if embedding_model was loaded on the vector store
+            if not embedding_model_name and hasattr(vector_store, 'embedding_model') and vector_store.embedding_model:
+                if hasattr(vector_store.embedding_model, 'model_name'):
+                    embedding_model_name = vector_store.embedding_model.model_name
+            
+            # Set session state
+            if embedding_model_name:
+                st.session_state.embedding_model = embedding_model_name
+            else:
+                st.session_state.embedding_model = None
+            
             # Display system prompt and model info
             provider_icon = "☁️" if llm_provider == "groq" else "🖥️"
             st.success(f"✅ Collection '{collection_name}' loaded successfully! {provider_icon} Using {llm_provider.upper()}")

vector_store.py

@@ -532,6 +532,48 @@ class QdrantManager:
         info = self.client.get_collection(collection_name)
         self.vector_size = info.config.params.vectors.size
         
+        # Try to load embedding model from first document's metadata
+        embedding_model_name = None
+        try:
+            # Scroll to get first point
+            points, _ = self.client.scroll(
+                collection_name=collection_name,
+                limit=1,
+                with_payload=True
+            )
+            if points and len(points) > 0:
+                payload = points[0].payload
+                embedding_model_name = payload.get("embedding_model")
+                if "chunking_strategy" in payload:
+                    self.chunking_strategy = payload["chunking_strategy"]
+        except Exception as e:
+            print(f"[QDRANT] Warning: Could not retrieve metadata: {e}")
+        
+        # If not found in metadata, try to infer from collection name
+        if not embedding_model_name:
+            # Collection name format: dataset_strategy_modelname
+            # Try common embedding models
+            known_models = [
+                "all-mpnet-base-v2",
+                "all-MiniLM-L6-v2", 
+                "paraphrase-MiniLM-L6-v2",
+                "multi-qa-MiniLM-L6-cos-v1"
+            ]
+            for model in known_models:
+                if model.lower().replace("-", "") in collection_name.lower().replace("-", "").replace("_", ""):
+                    embedding_model_name = f"sentence-transformers/{model}"
+                    break
+            # Default fallback
+            if not embedding_model_name:
+                embedding_model_name = "sentence-transformers/all-mpnet-base-v2"
+                print(f"[QDRANT] Warning: Could not determine embedding model, using default: {embedding_model_name}")
+        
+        # Load the embedding model
+        if embedding_model_name:
+            self.embedding_model = EmbeddingFactory.create_embedding_model(embedding_model_name)
+            self.embedding_model.load_model()
+            print(f"[QDRANT] Loaded embedding model: {embedding_model_name}")
+        
         print(f"[QDRANT] Loaded collection: {collection_name}")
         return self.current_collection
     
@@ -666,7 +708,9 @@ class QdrantManager:
                         "question": sample.get("question", ""),
                         "answer": sample.get("answer", ""),
                         "dataset": sample.get("dataset", ""),
-                        "total_docs": len(documents)
+                        "total_docs": len(documents),
+                        "embedding_model": embedding_model_name,
+                        "chunking_strategy": chunking_strategy
                     })
         
         # Add all chunks to collection