Set restrictions

app/rag.py

@@ -15,7 +15,14 @@ class ChatPDF:
     chain = None
 
     def __init__(self):
-        self.model = ChatOllama(model="qwen:1.8b")
+        self.model = ChatOllama(
+            model="qwen:1.8b",
+            keep_alive=-1,
+            temperature=0,
+            num_predict=512,
+            repeat_penalty=1.3,
+        )
+
         self.text_splitter = RecursiveCharacterTextSplitter(chunk_size=2048, chunk_overlap=128)
         self.prompt = PromptTemplate.from_template(
             """
@@ -37,7 +44,7 @@ class ChatPDF:
         self.retriever = vector_store.as_retriever(
             search_type="similarity_score_threshold",
             search_kwargs={
-                "k": 8,
+                "k": 4,
                 "score_threshold": 0.5,
             },
         )