Update src/simple_rag.py

src/simple_rag.py

@@ -103,10 +103,11 @@ def ask_question(query_text: str, k: int = 3):
         })
 
     context_text = "\n\n".join(chunk["text"] for chunk in context_chunks)
-    prompt = PROMPT_TEMPLATE.format(context=context_text, question=query_text).strip()
+    prompt = PROMPT_TEMPLATE.format(context=context_text, question=query_text)
     logging.info(f"Prompt: {prompt}")
     
     messages = [{"role": "user", "content": prompt}]
+    
     logging.info("Sending prompt to model...")
     logging.info(f"Messages: {messages}")
     
@@ -131,6 +132,7 @@ def ask_question(query_text: str, k: int = 3):
         max_new_tokens=256,
         return_full_text=False,
         truncation=True,
+        do_sample=False,
     )