RuntimeError: FlashAttention only supports Ampere GPUs or newer.

[nizhidao]

OS：Windows，GPU：2080ti 22g

is happened at when I trying to generate, is can not runing on my old turing?😢

can it be disabled? I can accept lower speed

[cjjkoko]

os.environ['ATTN_BACKEND'] = 'xformers'

[nizhidao]

os.environ['ATTN_BACKEND'] = 'xformers'

thanks, but it has a new problem: ModuleNotFoundError: No module named 'diff_gaussian_rasterization'

it happened when it generating, at randering process, how i can install this dependence?

[cjjkoko]

pip install kaolin==0.17.0 -f https://nvidia-kaolin.s3.us-east-2.amazonaws.com/torch-2.5.1_cu121.html

[nizhidao]

Duplicate of #26 Found same Issues, my cuda version is 12.7, should I lower to 11.8?

[cjjkoko]

Duplicate of #26 Found same Issues, my cuda version is 12.7, should I lower to 11.8?

https://kaolin.readthedocs.io/en/latest/notes/installation.html

[cjjkoko]

Maybe, cu121 is a good choice