Update rag.py

rag.py

@@ -34,7 +34,7 @@ loader = DirectoryLoader('.', glob="all_dialogues.txt")
 docs = loader.load()
 
 text_splitter = RecursiveCharacterTextSplitter(
-    chunk_size=1000, chunk_overlap=200
+    chunk_size=300, chunk_overlap=100
 )
 texts = text_splitter.split_documents(docs)
 embeddings = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")
@@ -49,7 +49,7 @@ db = FAISS.load_local(
 
 
 # Vector Store Retriever
-vector_retriever = db.as_retriever(search_kwargs={"k": 5})
+vector_retriever = db.as_retriever(search_kwargs={"k": 10})
 
 # Keyword Retriever (BM25)
 bm25_retriever = BM25Retriever.from_documents(texts)
@@ -76,24 +76,20 @@ def respond_rag_huggingface(message: str):
     context = "\n\n".join(doc.page_content for doc in docs)
 
     system_message = os.environ.get("SYSTEM_MESSAGE", 
-        "You are a Game of Thrones measter and harry potter's dumbledore"+
-        ". Answer the given question based on the context understand the context and try to answer based upon the context.do not give faulty answers and if any dialogue or description mention them ")
+        "You are a Game of Thrones maester and Harry Potter's Dumbledore. " +
+        "Answer the given question based on your knowledge, providing accurate details without mentioning any specific sources or context used. " +
+        "State how much you know about the topic, and do not provide faulty answers. " +
+        "If the answer is unclear, clarify what you mean rather than saying 'I do not know.'")
 
     prompt = f"""{system_message}
-
-Context:
-{context}
-
 Question:
 {message}
-
 Rules:
-- If the answer isn't in the context, respond with something you know about them and do not mention about about context
-- try to say that i exactly mean different than i do not know
+- Do not mention the context or where the information comes from
+- State how much you know about the topic (e.g., 'I have detailed knowledge,' 'I have some knowledge,' or 'My knowledge is limited')
 - Keep answers under 5 sentences
 - Include book/season references when possible
-- vector store contains both game of thrones and harry potter vectors so answer based on relevant context
-
+- Answer based on relevant knowledge from Game of Thrones and Harry Potter
 """
 
     response = model.generate_content(prompt)