Spaces:

multimodalart

/

stable-cascade

like 1.5k

Running on Zero

I see this app is using torch.compile for prior_pipeline and decoder_pipeline's unet. It works fine for my prior_pipeline.prior but decoder_pipeline.decoder failed. Does anyone know why?

decoder.decoder = torch.compile(decoder.decoder, mode="reduce-overhead", fullgraph=True)

`0%| | 0/10 [00:00<?, ?it/s][2024-02-14 01:51:43,802] [0/1] torch._subclasses.fake_tensor: [ERROR] fake tensor raised TypeError

[2024-02-14 01:51:43,802] [0/1] torch._subclasses.fake_tensor: [ERROR] TypeError: Multiple dispatch failed for 'torch.ops.aten.size'; all torch_dispatch handlers returned NotImplemented:
[2024-02-14 01:51:43,802] [0/1] torch._subclasses.fake_tensor: [ERROR]
[2024-02-14 01:51:43,802] [0/1] torch._subclasses.fake_tensor: [ERROR] - tensor subclass <class 'torch._subclasses.fake_tensor.FakeTensor'>
[2024-02-14 01:51:43,802] [0/1] torch._subclasses.fake_tensor: [ERROR]
[2024-02-14 01:51:43,802] [0/1] torch._subclasses.fake_tensor: [ERROR] For more information, try re-running with TORCH_LOGS=not_implemented`