Skip to content

fix(mhc): MHCPreNormFn.backward returns wrong number of gradients #10

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.

Sign up for GitHub

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

Open
yurekami wants to merge 1 commit into
base: main
Choose a base branch
from
Open

fix(mhc): MHCPreNormFn.backward returns wrong number of gradients #10

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 3 additions & 3 deletions tile_kernels/modeling/mhc/ops/norm_fn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -129,7 +129,7 @@ def forward(
def backward(
ctx: 'MHCPreNormFn',
out_grad: torch.Tensor,
) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor, None, None]:
) -> tuple[torch.Tensor | None, torch.Tensor, None, None, None]:
x, fn, out_mul, sqrsum = ctx.saved_tensors
norm_eps = ctx.norm_eps

Expand Down Expand Up @@ -166,8 +166,8 @@ def backward(

if ctx.fuse_grad_acc:
del x.untyped_storage().grad_from_mhc_post
return None, fn_grad, None, None, None, None
return x_grad, fn_grad, None, None, None, None
return None, fn_grad, None, None, None
return x_grad, fn_grad, None, None, None


def mhc_pre_norm_fn(
Expand Down