Skip to content
This repository has been archived by the owner on Aug 30, 2024. It is now read-only.

Sync jblas #18

Merged
merged 2 commits into from
Dec 29, 2023
Merged

Sync jblas #18

merged 2 commits into from
Dec 29, 2023

Conversation

luoyu-intel
Copy link
Contributor

Type of Change

Replace WeightKBlockF4 and WeightKBlockF8 with WeightKBlockNFloat.
Update PE core cache optimization.

@luoyu-intel
Copy link
Contributor Author

P/E hybrid client CPU performs poorly on NF4 and FP4 weight, due to unsupported hardware instruction vgatherdps on E cores. We would recommend that customers only use INT4 weight on these CPUs.

@kevinintel kevinintel merged commit b330746 into main Dec 29, 2023
9 checks passed
@luoyu-intel luoyu-intel deleted the sync_jblas branch December 29, 2023 05:57
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

5 participants