Skip to content

Conversation

@a-sidorova
Copy link
Contributor

@a-sidorova a-sidorova commented Dec 25, 2025

The third part of #45

@a-sidorova a-sidorova force-pushed the feature/layer_norm/asm branch 3 times, most recently from e32ce84 to aa92aca Compare December 26, 2025 09:14
@a-sidorova a-sidorova linked an issue Dec 26, 2025 that may be closed by this pull request
4 tasks
@a-sidorova a-sidorova mentioned this pull request Dec 26, 2025
4 tasks
@a-sidorova a-sidorova force-pushed the feature/layer_norm/asm branch from aa92aca to 482a7b5 Compare December 30, 2025 07:55
Signed-off-by: Alexandra Sidorova <[email protected]>

[Fusilli] Removed epsilon from the node signature

Signed-off-by: Alexandra Sidorova <[email protected]>
@a-sidorova a-sidorova force-pushed the feature/layer_norm/asm branch 2 times, most recently from 24c7cec to 2b09f3c Compare January 9, 2026 06:49
@a-sidorova a-sidorova marked this pull request as ready for review January 9, 2026 07:17
Comment on lines +245 to +253
// Scalar constants are inlined in the IR and not passed as function
// arguments, so they should not be in the variantPack.
if (input->isScalar()) {
FUSILLI_RETURN_ERROR_IF(variantPack.contains(input),
ErrorCode::VariantPackError,
"Scalar constant tensor found in variantPack");
continue;
}

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is this true for all scalars? This seems to be the case for epsilon based on Cudnn's layer norm sample.

For example, if you have a pointwise multiply X * s where s is a scalar value. This would have to be passed in through the variant pack.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[Fusilli] Support Layernorm

2 participants