Update app.py

app.py

@@ -225,25 +225,54 @@ prompt_template = ChatPromptTemplate.from_messages([
     ("human", "{question}")
 ])
 
-def process_question(question: str) -> tuple[str, str]:
-    # Check cache first
-    if question in question_cache:
-        return question_cache[question]
+# def process_question(question: str) -> tuple[str, str]:
+#     # Check cache first
+#     if question in question_cache:
+#         return question_cache[question]
     
-    relevant_docs = retriever(question)
-    context = "\n".join([doc.page_content for doc in relevant_docs])
+#     relevant_docs = retriever(question)
+#     context = "\n".join([doc.page_content for doc in relevant_docs])
     
-    prompt = prompt_template.format_messages(
-        context=context,
-        question=question
-    )
+#     prompt = prompt_template.format_messages(
+#         context=context,
+#         question=question
+#     )
     
-    response = llm(prompt)
-    result = (response.content, context)
+#     response = llm(prompt)
+#     result = (response.content, context)
     
-    # Cache the result
-    question_cache[question] = result
-    return result
+#     # Cache the result
+#     question_cache[question] = result
+#     return result
+
+def process_question(question: str):  
+    """  
+    Process the question and yield the answer progressively.  
+    """  
+    # Check cache first  
+    if question in question_cache:  
+        return question_cache[question]  # Retourne directement depuis le cache si disponible  
+
+    relevant_docs = retriever(question)  
+    context = "\n".join([doc.page_content for doc in relevant_docs])  
+
+    prompt = prompt_template.format_messages(  
+        context=context,  
+        question=question  
+    )  
+
+    response = ""  # Initialise la réponse  
+    # Ici, nous supposons que 'llm.stream' est un générateur qui renvoie des chunks  
+    for chunk in llm.stream(prompt):  # suppose que llm.stream renvoie des chunks de réponse  
+        if isinstance(chunk, str):  
+            response += chunk  # Accumulez la réponse si c'est déjà une chaîne  
+        else:  
+            response += chunk.content  # Sinon, prenez le contenu du chunk (si chunk est un type d'objet spécifique)  
+
+        yield response, context  # Renvoie la réponse mise à jour et le contexte  
+
+    # Mettez le résultat en cache à la fin  
+    question_cache[question] = (response, context) 
 
 # Custom CSS for right-aligned text in textboxes
 custom_css = """