update chunck

app/configs/prompts.py

@@ -1,6 +1,6 @@
 SYSTEM_PROMPTS = {
     # Supporto allo studio
-    "tutor": "Sei un tutor didattico di nome Valter. Usa questo contesto per rispondere: {context}",
+    "tutor": "Sei un tutor didattico di nome Valter. Usa questo contesto per rispondere: {context}. Cita sempre il titolo e l'autore dei documenti da cui prendi le informazioni. Inoltre, ricorda sempre di andare ad approfondire l'argomento",
     
     # Analisi dati scientifici
     "scientist": "Sei uno scienziato esperto. Analizza il contesto: {context}",

app/llm_handling.py

@@ -10,12 +10,16 @@ import gradio as gr
 import asyncio
 import edge_tts
 from pathlib import Path
+import requests
+from tenacity import retry, stop_after_attempt, wait_exponential
+import json
 
 from app.config import OPENAI_API_KEY
 from app.functions.database_handling import BASE_DB_PATH  # Aggiungi questo import
 from app.configs.prompts import SYSTEM_PROMPTS
 
 logging.basicConfig(level=logging.INFO)
+local_ip="192.168.82.5:1234"
 
 class LLMType(Enum):
     OPENAI_GPT_4O_MINI = "openai - GPT-4o-mini"
@@ -30,14 +34,14 @@ LLM_CONFIGS = {
         "base_url": None
     },
     LLMType.LOCAL_QWEN: {
-        "client": lambda: OpenAI(base_url="http://192.168.140.5:1234/v1", api_key="not-needed"),
+        "client": lambda: OpenAI(base_url="http://192.168.82.5:1234/v1", api_key="not-needed"),
         "model": "qwen2.5-coder-7b-instruct",
-        "base_url": "http://192.168.140.5:1234/v1"
+        "base_url": "http://192.168.82.5:1234/v1"
     },
     LLMType.LOCAL_PHI: {
-        "client": lambda: OpenAI(base_url="http://192.168.140.5:1234/v1", api_key="not-needed"),
+        "client": lambda: OpenAI(base_url="http://192.168.82.5:1234/v1", api_key="not-needed"),
         "model": "phi-3.5-mini-ita",
-        "base_url": "http://192.168.140.5:1234/v1"
+        "base_url": "http://192.168.82.5:1234/v1"
     }
 }
 
@@ -83,7 +87,7 @@ def clean_markdown(text):
     text = re.sub(r'```[\s\S]*?```', '', text)  # blocchi codice
     text = re.sub(r'`.*?`', '', text)           # codice inline
     text = re.sub(r'\[([^\]]+)\]\([^\)]+\)', r'\1', text)  # link
-    text = re.sub(r'\*\*(.*?)\*\*', r'\1', text)  # bold
+    text = re.sub(r'\*\*(.*?)\*\*\*', r'\1', text)  # bold
     text = re.sub(r'\*(.*?)\*', r'\1', text)      # italic
     return text.strip()
 
@@ -130,6 +134,15 @@ def get_system_prompt(prompt_type="tutor"):
     """Seleziona il prompt di sistema appropriato"""
     return SYSTEM_PROMPTS.get(prompt_type, SYSTEM_PROMPTS["tutor"])
 
+def test_local_connection():
+    """Verifica la connessione al server LLM locale"""
+    try:
+        response = requests.get(f"http://192.168.82.5:1234/v1/health", timeout=5)
+        return response.status_code == 200
+    except:
+        return False
+
+@retry(stop=stop_after_attempt(3), wait=wait_exponential(multiplier=1, min=4, max=10))
 def answer_question(question, db_name, prompt_type="tutor", chat_history=None, llm_type=LLMType.OPENAI_GPT_4O_MINI):
     """
     Risponde alla domanda 'question' usando i documenti del database 'db_name'.
@@ -163,11 +176,33 @@ def answer_question(question, db_name, prompt_type="tutor", chat_history=None, l
         vectorstore = FAISS.load_local(db_path, embeddings, allow_dangerous_deserialization=True)
 
         # Cerca i documenti (chunk) più simili
-        relevant_docs = vectorstore.similarity_search(question, k=3)
+        relevant_docs = vectorstore.similarity_search(question, k=5)
+
+        # Leggi il file metadata.json
+        metadata_file = os.path.join(db_path, "metadata.json")
+        metadata_dict = {}
+        if os.path.exists(metadata_file):
+            with open(metadata_file, 'r') as f:
+                metadata_list = json.load(f)
+                # Crea un dizionario per lookup veloce usando il filename come chiave
+                metadata_dict = {m["filename"]: m for m in metadata_list}
 
-        # Logga i chunk recuperati
+        # Logga i chunk recuperati con i loro metadata
         for idx, doc in enumerate(relevant_docs):
             logging.info(f"--- Chunk {idx+1} ---")
+            # Recupera i metadata dal documento
+            source_file = doc.metadata.get("source", "Unknown")
+            chunk_info = f"File: {source_file}"
+            
+            # Aggiungi informazioni dal metadata.json se disponibili
+            if source_file in metadata_dict:
+                file_metadata = metadata_dict[source_file]
+                chunk_info += f"\nTitolo: {file_metadata['title']}"
+                chunk_info += f"\nAutore: {file_metadata['author']}"
+                chunk_info += f"\nData caricamento: {file_metadata['upload_date']}"
+            
+            logging.info(chunk_info)
+            logging.info("Contenuto:")
             logging.info(doc.page_content)
             logging.info("---------------------")
 
@@ -190,6 +225,10 @@ def answer_question(question, db_name, prompt_type="tutor", chat_history=None, l
             {"role": "user", "content": question}
         ]
         
+        if "local" in str(llm_type):
+            if not test_local_connection():
+                raise ConnectionError("LM Studio non raggiungibile")
+                
         client, model = get_llm_client(llm_type)
         response = client.chat.completions.create(
             model=model,
@@ -210,9 +249,10 @@ def answer_question(question, db_name, prompt_type="tutor", chat_history=None, l
 
     except Exception as e:
         logging.error(f"Errore durante la generazione della risposta: {e}")
+        error_msg = "LLM locale non disponibile. Riprova più tardi o usa OpenAI." if "local" in str(llm_type) else str(e)
         return [
             {"role": "user", "content": question},
-            {"role": "assistant", "content": f"Si è verificato un errore: {str(e)}"}
+            {"role": "assistant", "content": f"⚠️ {error_msg}"}
         ]
 
 

ui/chatbot_tab.py

@@ -251,3 +251,4 @@ def create_chatbot_tab():
 
     # Ritorna il riferimento al dropdown corretto
     return {"db_selector": db_name_chat}
+