Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

flash attention not compatible? #20

Open
oxPJ opened this issue Aug 25, 2024 · 1 comment
Open

flash attention not compatible? #20

oxPJ opened this issue Aug 25, 2024 · 1 comment

Comments

@oxPJ
Copy link

oxPJ commented Aug 25, 2024

When I try to train a stripedhyena model I keep getting issues with the stripedhyena modules seemingly trying to import modules from Flash Attention in an outdated way.
example:
AttributeError: module 'dropout_layer_norm' has no attribute 'dropout_add_ln_fwd'

For some modules, I could solve this by creating mock classes that redirect the module import to the correct modules within the current flash-attention implementation. But others (like 'dropout_add_ln_fwd') I can't find in flash-attn at all.

on a related note the current repo seems to be of StripedHyena 0.2.1 while the newest implementation is 0.2.2? (but issue presists on both 0.2.2 and 0.2.1)

Could you provide guidance on how to resolve these import issues or update the repository to ensure compatibility with the latest Flash Attention modules?

Thank you!

@sun-qibo
Copy link

sun-qibo commented Aug 28, 2024

I also had the error for module 'dropout_layer_norm'. resolved by
pip install git+https://github.com/HazyResearch/flash-attention.git#subdirectory=csrc/layer_norm

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants