enable_xformers_memory_efficient_attention is not supported

[JarvisFei]

File "fastcomposer/fastcomposer/model.py", line 571, in forward
localization_loss = get_object_localization_loss(
File "fastcomposer/model.py", line 416, in get_object_localization_loss
return loss / num_layers
ZeroDivisionError: division by zero

May I ask if there is a solution to this problem when i am using enable_xformers_memory_efficient_attention？
@Guangxuan-Xiao

[JarvisFei]

in the model.py file, why the code is like this:

if isinstance(module.processor, AttnProcessor2_0): module.set_processor(AttnProcessor())

how can I accelerate the training process with torch.complie() when I am using PyTorch 2.0

[xilanhua12138]

same problem, did you solve it ? can you share?