Update to phi3:mini

app/rag.py

@@ -15,11 +15,11 @@ class ChatPDF:
     chain = None
 
     def __init__(self):
-        self.model = ChatOllama(model="gemma:2b")
+        self.model = ChatOllama(model="phi3:mini")
         self.text_splitter = RecursiveCharacterTextSplitter(chunk_size=128, chunk_overlap=8)
         self.prompt = PromptTemplate.from_template(
             """
-            <s> [INST] Use the pieces of context to answer the question concisely.[/INST] </s> 
+            <s> [INST] Use the context to answer the question.[/INST] </s> 
             [INST] Question: {question} 
             Context: {context} 
             Answer: [/INST]
@@ -35,7 +35,7 @@ class ChatPDF:
         self.retriever = vector_store.as_retriever(
             search_type="similarity_score_threshold",
             search_kwargs={
-                "k": 56,
+                "k": 24,
                 "score_threshold": 0.5,
             },
         )

start_service.sh

@@ -7,7 +7,7 @@ ollama serve &
 sleep 5
 
 # Pull and run <YOUR_MODEL_NAME>
-ollama pull gemma:2b
+ollama pull phi3:mini
 
 #
 fastapi run /code/app/main.py --port 7860