Warning message: Enabling CPU offloading option for models

[ajikmr]

File: model_init.py

pipe.enable_model_cpu_offload() return pipe.to(device)

It seems after enabling CPU offloading option, model is send to CUDA device. It is done so in a number of model initializations. It seems the correct option would be:

pipe.enable_model_cpu_offload() return pipe

For cases, where model offloading is not done: return pipe.to(device)

Thanks,