CUDA

src/pipeline.py

@@ -108,7 +108,7 @@ def load_pipeline() -> FluxPipeline:
                                         text_encoder=text_encoder,
                                         text_encoder_2=text_encoder_2,
                                         torch_dtype=torch.bfloat16).to("cuda")
-    pipe.enable_sequential_cpu_offload()
+    # pipe.enable_sequential_cpu_offload()
     pipe("cat", guidance_scale=0., max_sequence_length=256, num_inference_steps=4)
     return pipe