Skip to content

Exclude RMSNormGated from 4bit weight quantization (#277) - #281

Merged
stikves merged 3 commits into
apple:mainfrom
stikves:sukru/fix-rmsnormgated-4bit-exclusion
Sep 23, 2026
Merged

stikves merged 3 commits into
apple:mainfrom
stikves:sukru/fix-rmsnormgated-4bit-exclusion

Conversation

@stikves

@stikves stikves commented Sep 22, 2026

Copy link
Copy Markdown
Contributor

RMSNormGated does not subclass RMSNorm, so it missed the name-based quantization exclusion and the global INT4 spec (per_block, axis=1) was applied to its rank-1 weight, raising "axis 1 out of bounds for tensor of rank 1". Add its module path to _TORCH_MODULE_EXCLUSIONS alongside the other RMSNorm composite variants.

stikves and others added 2 commits September 22, 2026 07:05
RMSNormGated does not subclass RMSNorm, so it missed the name-based
quantization exclusion and the global INT4 spec (per_block, axis=1) was
applied to its rank-1 weight, raising "axis 1 out of bounds for tensor
of rank 1". Add its module path to _TORCH_MODULE_EXCLUSIONS alongside the
other RMSNorm composite variants.
@stikves
stikves marked this pull request as ready for review September 23, 2026 01:03
@stikves stikves self-assigned this Sep 23, 2026
Comment thread python/tests/test_model_units/test_primitives/test_macos/test_rms_norm.py Outdated
…rms_norm.py

Co-authored-by: Prathamesh <46148373+pkmandke@users.noreply.github.com>
@stikves
stikves merged commit f7d42b1 into apple:main Sep 23, 2026
3 checks passed
@stikves
stikves deleted the sukru/fix-rmsnormgated-4bit-exclusion branch September 23, 2026 03:54
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.

2 participants