Update app.py

app.py

@@ -68,8 +68,8 @@ async def lifespan(app: FastAPI):
         n_ctx=4096,
         
         # Threading optimizations
-        n_threads=8,       
-        n_threads_batch=8, 
+        n_threads=4,       
+        n_threads_batch=4, 
         numa=NUMA_STRATEGY,
         
         # ⚡ TURBOQUANT EQUIVALENT SETTINGS: