torch.Generator is not supported on Intel so we have to run BrownianTreeNoiseSampler on the CPU with transform and send results back to the GPU with transform_last.
It will throw this error if we try sending it to the GPU outside of the class:
AttributeError: 'BrownianTreeNoiseSampler' object has no attribute 'to'
torch.Generator is not supported on Intel so we have to run BrownianTreeNoiseSampler on the CPU with transform and send results back to the GPU with transform_last.
It will throw this error if we try sending it to the GPU outside of the class: AttributeError: 'BrownianTreeNoiseSampler' object has no attribute 'to'
Example use: return BrownianTreeNoiseSampler(x.to("cpu"), sigma_min, sigma_max, seed=current_iter_seeds, transform=lambda x: x.to("cpu"), transform_last=lambda x: x.to("xpu"))