Add CPU fallback for latent upscale OOM

dumas_h3_latent_upscale.py

@@ -602,6 +602,20 @@ def upscale_video_model(video, param):
             raise
         smaller = _shrink_model_tile_param(param)
         if smaller is None:
+            if str(param.get("device", "cuda")) != "cpu":
+                cpu_param = dict(param)
+                cpu_param["device"] = "cpu"
+                cpu_param["precision"] = "fp32"
+                try:
+                    gc.collect()
+                except Exception:
+                    pass
+                if torch.cuda.is_available():
+                    try:
+                        torch.cuda.empty_cache()
+                    except Exception:
+                        pass
+                return upscale_video_model(video, cpu_param)
             raise
         try:
             gc.collect()