Update app.py

app.py

@@ -11,6 +11,8 @@ try:
         device_map="cpu",
         torch_dtype=torch.float16,  # Use float16 for reduced memory
         low_cpu_mem_usage=True,     # Enable memory optimization
+        max_memory={"cpu": "15GB"},  # Limit memory usage
+        offload_folder="offload",    # Enable disk offloading
     )
     
     # Load tokenizer