fix_ddpo_demo

[zhangsibo1129]

In examples/scripts/ddpo.py , the AestheticScorer.mlp loads weights to cuda device by default

self.mlp.load_state_dict(state_dict)
state_dict = torch.load(cached_path)

which throws an error for non-cuda devices, so loading weights on cpu first

self.mlp.load_state_dict(state_dict, map_location=torch.device('cpu'))
state_dict = torch.load(cached_path)

and then copy to target device

if is_npu_available():
    scorer = scorer.npu()
elif is_xpu_available():
    scorer = scorer.xpu()
else:
    scorer = scorer.cuda()

[younesbelkada]

Thanks for fixing the ddpo demo !

[younesbelkada]

@zhangsibo1129 can you run the styling checks? make precommit

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[younesbelkada]

Thanks again!