Re:Hidet: A Deep Learning Compiler for Efficient Model Serving
by mlazos on Friday April 28, 2023 @01:20AM
(#35737696)
This bug was in the front end (dynamo) so I’m not surprised it happens with hidet too. I fixed this for the 2.0.1 release.
Re:Hidet: A Deep Learning Compiler for Efficient Model Serving
by brucethemoose2 on Friday April 28, 2023 @01:39AM
(#35737817)
I am getting backend errors right now. For instance, with the unet:
> torch._dynamo.exc.BackendCompilerFailed: backend='hidet_backend' raised: NotImplementedError: hidet: Tensor.to(..., device=...) is not supported for symbolic tensors...
Early torch 2.0 was full of backend unimplemented/unsupported errors too, but I can't even remember when dynamo issues popped up (or even when everything started working) since I stayed on torch nightly.
Re:Hidet: A Deep Learning Compiler for Efficient Model Serving
by mlazos on Friday April 28, 2023 @01:56AM
(#35737921)
Ah ok this wasn’t what I thought then. I haven’t taken a look at Hidet yet so I won’t be much help here :( I thought you were talking about 2.0.0
Re:Hidet: A Deep Learning Compiler for Efficient Model Serving
by yyding on Friday April 28, 2023 @02:21AM
(#35738075)
Hi, do you mind submitting an issue to the hidet github repo so we can have a look? Thank you!
Re:Hidet: A Deep Learning Compiler for Efficient Model Serving
by akbarnur on Friday April 28, 2023 @09:36AM
(#35741115)
Hey @bructhemoose2 can you file an issue, we will try to fix it ASAP: https://github.com/hidet-org/hidet/issues
But inductor/triton didn't work in the 2.0 nightlies either, and now it works fine for SD.