Update app.py

app.py

@@ -66,7 +66,7 @@ def load_bf16_pipeline():
         end_time = time.time()
         mem_reserved = torch.cuda.memory_reserved(0)/1024**3 if DEVICE == "cuda" else 0
         print(f"BF16 Pipeline loaded in {end_time - start_time:.2f}s. Memory reserved: {mem_reserved:.2f} GB")
-        # CACHED_PIPES[MODEL_ID] = pipe
+        CACHED_PIPES[MODEL_ID] = pipe
         return pipe
     except Exception as e:
         print(f"Error loading BF16 pipeline: {e}")