Reduce tokens for faster CPU inference

app/ora_server.py

@@ -98,7 +98,7 @@ async def chat_endpoint(req: ChatRequest):
 
     outputs = model.generate(
         input_ids,
-        max_new_tokens=256,
+        max_new_tokens=128,  # Reduced for faster CPU inference
         eos_token_id=terminators,
         do_sample=True,
         temperature=0.7,