Update to qwen:1.8b

app/rag.py

@@ -15,7 +15,7 @@ class ChatPDF:
     chain = None
 
     def __init__(self):
-        self.model = ChatOllama(model="qwen:0.5b")
+        self.model = ChatOllama(model="qwen:1.8b")
         self.text_splitter = RecursiveCharacterTextSplitter(chunk_size=2048, chunk_overlap=128)
         self.prompt = PromptTemplate.from_template(
             """
@@ -37,7 +37,7 @@ class ChatPDF:
             search_type="similarity_score_threshold",
             search_kwargs={
                 "k": 15,
-                "score_threshold": 1.8,
+                "score_threshold": 0.5,
             },
         )
 

start_service.sh

@@ -7,7 +7,7 @@ ollama serve &
 sleep 5
 
 # Pull and run <YOUR_MODEL_NAME>
-ollama pull qwen:0.5b
+ollama pull qwen:1.8b
 
 #
 fastapi run /code/app/main.py --port 7860