Update app.py

app.py

@@ -216,41 +216,46 @@ class RAGPipeline:
         self.retriever = SentenceTransformerRetriever()
         self.documents = []
         self.device = torch.device("cpu")
-        self.model_path = os.path.join("models", "mistral-7b-v0.1.Q4_K_M.gguf")
+        
+        # Change 1: Process documents first
+        self.load_and_process_csvs()
+        
+        # Change 2: Simplified model path
+        self.model_path = "mistral-7b-v0.1.Q4_K_M.gguf"
         self.llm = None
+        # Change 3: Initialize model after documents are processed
         self._initialize_model()
 
-    @st.cache_resource(show_spinner=False)
+    @st.cache_resource  # Added caching decorator
     def _initialize_model(_self):
         try:
-            os.makedirs(os.path.dirname(_self.model_path), exist_ok=True)
-            
             if not os.path.exists(_self.model_path):
                 direct_url = "https://huggingface.co/TheBloke/Mistral-7B-v0.1-GGUF/resolve/main/mistral-7b-v0.1.Q4_K_M.gguf"
                 download_file_with_progress(direct_url, _self.model_path)
 
+            # Added better error handling
             if not os.path.exists(_self.model_path):
                 raise FileNotFoundError(f"Model file {_self.model_path} not found after download attempts")
-            
-            if os.path.getsize(_self.model_path) < 1000000:
-                os.remove(_self.model_path)
-                raise ValueError("Downloaded model file is too small, likely corrupted")
-            
+        
+            # Added verbose mode for better debugging
             llm_config = {
                 "n_ctx": 2048,
                 "n_threads": 4,
                 "n_batch": 512,
                 "n_gpu_layers": 0,
-                "verbose": False
+                "verbose": True  # Added this
             }
-        
+    
             _self.llm = Llama(model_path=_self.model_path, **llm_config)
             st.success("Model loaded successfully!")
+        
         except Exception as e:
+            # Added better error logging
             logging.error(f"Error initializing model: {str(e)}")
             st.error(f"Error initializing model: {str(e)}")
             raise
-
+    
+        
     def check_model_health(self):
         try:
             if self.llm is None: