Try sequential CPU offload for even more memory savings

app.py

@@ -66,7 +66,11 @@ log_memory("After loading pipeline")
 # This keeps model weights on CPU and only moves to GPU when needed
 log_step("Enabling CPU offloading for memory efficiency")
 try:
-    if hasattr(pipe, 'enable_model_cpu_offload'):
+    # Try sequential CPU offload first (most memory efficient)
+    if hasattr(pipe, 'enable_sequential_cpu_offload'):
+        pipe.enable_sequential_cpu_offload()
+        print("Enabled sequential CPU offload (most memory efficient)")
+    elif hasattr(pipe, 'enable_model_cpu_offload'):
         pipe.enable_model_cpu_offload()
         print("Enabled model CPU offload")
     else: