Cuda: Decoder Masked Multihead Attention Q values get corrupted when using cross attention #16721
This was accidentally moved into the `if(!params.is_cross_attention)`…
33c94854
gh-yewang
approved these changes
on 2023-07-14
RyanUnderhill
deleted the ryanunderhill/mmha_q_fix branch 3 years ago
Assignees
No one assigned
Login to write a write a comment.
Login via GitHub