Skip to content

Remove deprecated block pointer from batched flash attention benchmark#7568

Draft
whitneywhtsang wants to merge 1 commit into
mainfrom
whitneywhtsang/batched_fa
Draft

Remove deprecated block pointer from batched flash attention benchmark#7568
whitneywhtsang wants to merge 1 commit into
mainfrom
whitneywhtsang/batched_fa

Conversation

@whitneywhtsang

Copy link
Copy Markdown
Contributor

The l (log-sum-exp) store used a rank-1 block pointer with non-unit stride, which has no legal tensor descriptor equivalent. Replace with a masked tensor-of-pointer store and remove the now-unused load_if and store_if helpers.

The l (log-sum-exp) store used a rank-1 block pointer with non-unit
stride, which has no legal tensor descriptor equivalent. Replace with
a masked tensor-of-pointer store and remove the now-unused load_if
and store_if helpers.

Signed-off-by: Whitney Tsang <whitney.tsang@intel.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants