Update app.py

app.py

@@ -18,7 +18,7 @@ OLLAMA_SERVICE_THREAD.start()
 
 print("Giving ollama serve a moment")
 time.sleep(10)
-subprocess.run("~/ollama pull gemma2:27b", shell=True)
+subprocess.run("~/ollama pull gemma2", shell=True)
 
 
 import copy
@@ -81,7 +81,7 @@ def stream_chat(message: str, history: list, temperature: float, context_window:
 
     llm = gemma2()
 
-    resp = llm.stream_chat(
+    resp = llm.chat(
         messages = messages,
         chat_history = conversation,
         top_p=top_p,
@@ -90,9 +90,7 @@ def stream_chat(message: str, history: list, temperature: float, context_window:
         context_window=context_window,            
     )
         
-
-    for r in resp:
-        yield r.delta
+    return resp
 
 
 chatbot = gr.Chatbot(height=600)