Roman Treutlein

Results 13 comments of Roman Treutlein

Mh for that usecase i think you should use OpenPsi and let it decide which goal/demand to fullfill. But you will also need have a module for OpenPsi. In regards...

I just tried with a fresh venv but have the same issue.

Okay i had to install the cuda toolkit to be able to build flash attention. I guess a note that cuda dependencies installed by torch alone are not enough (as...