consistency_models icon indicating copy to clipboard operation
consistency_models copied to clipboard

torch version

Open XipengY opened this issue 1 year ago • 5 comments

Hi, Thanks for your interesting work first! I am confused about the environment when running the code, and want to know the versions of python and torch used in your project.

Thanks very much!

XipengY avatar Apr 15 '23 17:04 XipengY

I was able to run this using Python 3.8.16, Torch 2.0.0. Additionally, I'm using Cuda version 11.0 and needed to use Cuda version 11.3 to do evaluation (see evaluations/ for separate compilation).

nfrumkin avatar Apr 15 '23 23:04 nfrumkin

Thanks for your replay!

I found torch==2.0.0 need CUDA>=11.7, we build torch by source? I use 'pip install -e .' the default version of flash-attn is 1.0.1, but I also found flash-attn==1.0.1 need CUDA>=11.4. So I'm also wondering how to resolve this conflict about the environment. Would it be convenient for you to share your project environment by docker or Anaconda?

Thanks very much.

XipengY avatar Apr 16 '23 13:04 XipengY

I use 'conda install pytorch torchvision cudatoolkit=11.7 -c pytorch', and pip install flash-attn==0.2.8, solved the environment.

XipengY avatar Apr 17 '23 03:04 XipengY

I use 'conda install pytorch torchvision cudatoolkit=11.7 -c pytorch', and pip install flash-attn==0.2.8, solved the environment.

i cant install 0.2.8 using pip install flash-attn==0.2.8, do you have this problem? image image

treefreq avatar Apr 22 '23 09:04 treefreq

try torch version 2.0.0?

XipengY avatar Apr 27 '23 08:04 XipengY