Stonepia
Stonepia
Added @Stonepia for product definition.
Hi @blue-notes-robot , thanks for the report! I could reproduce it on my local machine. I believe it should be a bug of the driver, and because the device (Gen...
Hi, sorry for the late reply. This issue is the same with https://github.com/pytorch/pytorch/issues/149953 This issue is because the iGPU is old and lacks of the fp64 support, but the PyTorch's...
This might be a known issue. When the system memory is under high pressure, it might cause this UR Error. You could witness that the system memory is almost full,...
Well, for SD1.5, then there is no reason it should fail with this error. SDv3 will have this UR Error due to the memory pressure. Could you take a look...
> > Could you take a look at how much memory it uses when the model runs? Seems that you are using Linux, you may use `xpu-smi` for example. >...
Hi @lostkingdom4 , thanks for the detailed reproducer! I did a check, the problem is because the registered meta function's schema does not find the exact custom op. Actually, if...
Hi @lostkingdom4 , I suspect this should be a bug by PyTorch, not your implementation issue. We haven't found it because no one has tried this path (custom ops registered...
Hi, @lostkingdom4 Yes, that's why I said that the problem is not your implementation, it should be PyTorch side issue (not even IPEX side). So I would assume to write...
I don't think you need to rebuild everytime, you could try first with the Python Op registration, it should be the same with C++ side. You don't need to build...