I use
pip install flash_attn-2.7.0.post2%2Bcu124torch2.4.0cxx11abiFALSE-cp311-cp311-win_amd64.whl
I want to know how to call it correctly—should I use:
from flash_attn import flash_attn_func
or
from torch.nn.functional import scaled_dot_product_attention
Additionally, I only installed the .whl file and did not install ninja. Is this correct?