We read every piece of feedback, and take your input very seriously.
To see all available qualifiers, see our documentation.
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
Q1: According to the DOC of cudnn graph API description for SDPA graph with Alibi mask: But actually, the code implementation has nothing to do with sliding windows: https://github.com/NVIDIA/cudnn-frontend/blob/1.7.0-rc/include/cudnn_frontend/node/scaled_dot_product_flash_attention.h#L300 I am just curious about how the Alibi mask is described in cudnn frontend API and how can I construct that with sub graph? It seems the graph pic in doc is mismatched with code.
Q2: Also, seems that the naming of the function get_abili_slope is wrong. ----> get_alibi_slope https://github.com/NVIDIA/cudnn-frontend/blob/1.7.0-rc/include/cudnn_frontend/node/scaled_dot_product_flash_attention.h#L670
get_abili_slope
get_alibi_slope
The text was updated successfully, but these errors were encountered:
No branches or pull requests
Q1:
According to the DOC of cudnn graph API description for SDPA graph with Alibi mask:
But actually, the code implementation has nothing to do with sliding windows:
https://github.com/NVIDIA/cudnn-frontend/blob/1.7.0-rc/include/cudnn_frontend/node/scaled_dot_product_flash_attention.h#L300
I am just curious about how the Alibi mask is described in cudnn frontend API and how can I construct that with sub graph? It seems the graph pic in doc is mismatched with code.
Q2:
Also, seems that the naming of the function
get_abili_slope
is wrong. ---->get_alibi_slope
https://github.com/NVIDIA/cudnn-frontend/blob/1.7.0-rc/include/cudnn_frontend/node/scaled_dot_product_flash_attention.h#L670
The text was updated successfully, but these errors were encountered: