search

castorini / daam

Diffusion attentive attribution maps for interpreting Stable Diffusion.
MIT License
669 stars 61 forks source link

RuntimeError: stack expects a non-empty TensorList #12 #25

Closed UdonDa closed 1 year ago

UdonDa commented 1 year ago

Hi, @daemon

I encounter the same error as #12 . My program is the same as the example code in https://github.com/castorini/daam#:~:text=as%20a%20Library-,Import,-and%20use%20DAAM .

I was debugging daam, but I do not know the reason. Do you know the reason?

Screenshot 2022-12-24 18 26 10
daemon commented 1 year ago

Curious, I'm looking into it. I wasn't able to reproduce the error on my end -- I'll try with a fresh install soon. Here's my environment: https://gist.github.com/daemon/4ffc94ab4454777cfedeb95d1b7c7d42

daemon commented 1 year ago

Did a fresh install on an AWS G5 instance with an Ubuntu Deep Learning AMI, wasn't able to reproduce either: https://gist.github.com/daemon/ba5d62ff65b19be5686e9c0081bf2748

It's possible the installed hook (see trace.py) has been overridden by something that's not part of DAAM. Are there any other packages that would have the same effect? Is a package forcing a different kind of attention to be used?

daemon commented 1 year ago

Please re-open if this is still a problem.