search

SAIS-FUXI / VidGen

55 stars 4 forks source link

Xformer conflict with torch2.2.2cuda118 #5

Open shikharRS opened 1 month ago

shikharRS commented 1 month ago

Hi team, i am not able to resolve the issue of version conflict. Can you help me with the wheel file for xformer , which can be compatible with torch2.2.2cuda118

WARNING[XFORMERS]: xFormers can't load C++/CUDA extensions. xFormers was built for: PyTorch 2.4.1+cu118 with CUDA 1108 (you have 2.2.2+cu118)

SAIS-FUXI commented 1 month ago

You can access it from this link（ https://download.pytorch.org/whl/xformers/ ）Find the appropriate version below to download and install. image

You can also install it directly through the following command: pip install -U xformers==0.0.25 --index-url https://download.pytorch.org/whl/cu118

shikharRS commented 1 month ago

Thanks , that issue is resolved but i am facing another challenge of OOM(out of Memory). I have 2 GPU each of 24GB approx .Below is the output of nvidia-smi

image

shikharRS commented 1 month ago

Do i require more computation resources ? 2 GPU of each having 24GB is not sufficient ?

shikharRS commented 1 month ago

Please let me know the computation resources require to execute the code

CanvaChen commented 1 month ago

@shikharRS 2 GPU (24GB + 12GB) should be sufficient. However, the code may require some optimization. You can refer to this fork https://github.com/CanvaChen/VidGen