Skip to content

feat: add linear_packed variants with pre-transposed weight - #1458

Merged
wooway777 merged 2 commits into
InfiniTensor:mainfrom
DreamStyle0919:pre-transpose
Aug 3, 2026
Merged

feat: add linear_packed variants with pre-transposed weight#1458
wooway777 merged 2 commits into
InfiniTensor:mainfrom
DreamStyle0919:pre-transpose

Conversation

@DreamStyle0919

Copy link
Copy Markdown

linear_packed() and linear_packed_() accept weight already in [IC, OC] layout, bypassing the permute({1, 0}) call inside linear_() to reduce runtime overhead when weights are pre-packed.

linear_packed() and linear_packed_() accept weight already in [IC, OC]
layout, bypassing the permute({1, 0}) call inside linear_() to reduce
runtime overhead when weights are pre-packed.
@DreamStyle0919
DreamStyle0919 requested a review from a team August 3, 2026 08:03
@wooway777
wooway777 merged commit 6620720 into InfiniTensor:main Aug 3, 2026
5 checks passed
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.

3 participants