-
Notifications
You must be signed in to change notification settings - Fork 39
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
Towards RPE #475
base: main
Are you sure you want to change the base?
Towards RPE #475
Conversation
iree/turbine/kernel/wave/codegen.py
Outdated
subs[-1] = ( | ||
subs[-1][0], | ||
start_indices_orig[-1] + idxc.iota(elements_per_thread), | ||
start_indices_orig[-1], # + idxc.iota(elements_per_thread), |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
IMPORTANT: iota is still needed when offset uses scalar / broadcasted symbols, but not needed when offset uses vector symbols.
Not sure of the repercussions on the design
@@ -1420,6 +1420,7 @@ def check_is_mapping_contiguous( | |||
return True | |||
|
|||
# TODO: Better dyn vals analysis. | |||
# TODO TODO TODO: we also need to check if there are additional sybols in the mapping |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
THis is done below
This implements (causal) ALiBi, attention with linear biases, following the paper "Train Short, Test Long: Attention with Linear Biases Enables Input Length Extrapolation" by Press et.al in ICLR 2022. Signed-off-by: Alex Zinenko <[email protected]>
Signed-off-by: Alex Zinenko <[email protected]>
Signed-off-by: Alex Zinenko <[email protected]>
Signed-off-by: Nicolas Vasilache <[email protected]>
3015755
to
a5a1818
Compare
No description provided.