Upload VideoEngine_optimized.py

VideoEngine_optimized.py

@@ -149,27 +149,12 @@ class VideoEngine:
                 print(f"⚠ Quantization failed: {e}")
                 raise RuntimeError("FP8 quantization required for this optimized version")
 
-            # Stage 4: Load AOTI blocks
-            print("→ [4/5] Loading AOTI blocks...")
-            try:
-                import aoti
-
-                aoti.aoti_blocks_load(
-                    self.pipeline.transformer,
-                    'zerogpu-aoti/Wan2',
-                    variant='fp8da'
-                )
-                aoti.aoti_blocks_load(
-                    self.pipeline.transformer_2,
-                    'zerogpu-aoti/Wan2',
-                    variant='fp8da'
-                )
-                print("✓ AOTI blocks loaded (1.5-1.8x speedup)")
-                self.use_aoti = True
-            except Exception as e:
-                print(f"⚠ AOTI loading failed: {e}")
-                print("  Continuing without AOTI (FP8 only)")
-                self.use_aoti = False
+            # Stage 4: AOTI blocks (currently disabled for stability)
+            print("→ [4/5] Skipping AOTI (using FP8 only for stability)...")
+            # AOTI can cause runtime errors with certain PyTorch versions
+            # FP8 quantization alone provides excellent performance
+            self.use_aoti = False
+            print("✓ Using FP8 quantization only (stable and fast)")
 
             # Stage 5: Move to GPU and enable optimizations
             print("→ [5/5] Moving to GPU...")