llama prompt and debug update

config/prompt_templates/llama2.txt

@@ -1,6 +1,7 @@
-<<SYS>>
-**CONTEXT FOR THE QUESTION**:
-{context}
+<<SYS>>Using the information contained in the context, 
+give a comprehensive answer to the question.
+Respond only to the question asked, response should be concise and relevant to the question.
+If the answer cannot be deduced from the context, do not give an answer.
 <</SYS>>
 
-[INST] Respond for the user question as factualy as possible, using given context. [/INST] User: {question}
+[INST] Context: {context} [/INST] User: {question}

qa_engine/qa_engine.py

@@ -227,7 +227,7 @@ class QAEngine():
             self.knowledge_index = FAISS.load_local('./indexes/run/', embedding_model)
             self.reranker = CrossEncoder('cross-encoder/ms-marco-MiniLM-L-12-v2')
 
-    
+
     @staticmethod
     def _preprocess_question(question: str) -> str:
         if question[-1] != '?':
@@ -300,8 +300,8 @@ class QAEngine():
         logger.info('Running LLM chain')
         question_processed = QAEngine._preprocess_question(question)
         answer = self.llm_chain.run(question=question_processed, context=context)
-        answer = QAEngine._postprocess_answer(answer)
-        response.set_answer(answer)
+        answer_postprocessed = QAEngine._postprocess_answer(answer)
+        response.set_answer(answer_postprocessed)
         logger.info('Received answer')
 
         if self.debug:
@@ -310,7 +310,8 @@ class QAEngine():
             logger.info(f'question len: {len(question)} {sep}')
             logger.info(f'question: {question} {sep}')
             logger.info(f'answer len: {len(response.get_answer())} {sep}')
-            logger.info(f'answer: {response.get_answer()} {sep}')
+            logger.info(f'answer original: {answer} {sep}')
+            logger.info(f'answer postprocessed: {response.get_answer()} {sep}')
             logger.info(f'{response.get_sources_as_text()} {sep}')
             logger.info(f'messages_contex: {messages_context} {sep}')
             logger.info(f'relevant_docs: {relevant_docs} {sep}')