Update skyreelsinfer/skyreels_video_infer.py

skyreelsinfer/skyreels_video_infer.py

@@ -47,10 +47,10 @@ class SkyReelsVideoSingleGpuInfer:
             device="cpu",
         ).to("cpu")
         if quant_model:
-            quantize_(text_encoder, float8_weight_only(), device=gpu_device)
+            quantize_(text_encoder, float8_weight_only(), device="cpu")
             text_encoder.to("cpu")
             torch.cuda.empty_cache()
-            quantize_(transformer, float8_weight_only(), device=gpu_device)
+            quantize_(transformer, float8_weight_only(), device="cpu")
             transformer.to("cpu")
             torch.cuda.empty_cache()
         pipe = SkyreelsVideoPipeline.from_pretrained(