onnxruntime
Enable KleidiAI for asymmetric 4-bit MatMulNBits on ARM64
#27751

Merged

Enable KleidiAI for asymmetric 4-bit MatMulNBits on ARM64 #27751

jambayk merged 10 commits into main from jambayk/mnb-arm-32

init

5adbcb3e

optimized correction

35e6c85c

use naive loop

f602a51d

neon correction

afc0ae20

fix correction

2d74cd75

more opt

30fe481b

jambayk requested a review from

copilot-pull-request-reviewer 13 days ago

copilot-pull-request-reviewer commented on 2026-03-18

Fix float alignment and per-entry zero point gating

f417a99b

jambayk requested a review from

copilot-pull-request-reviewer 13 days ago

copilot-pull-request-reviewer commented on 2026-03-19

Address review: strict aliasing, dynamic zero_points guard, doc

a25adc7d

Fix fp16 path missing BZpCorr on macOS ARM

6684be5e

jambayk enabled auto-merge (squash) 13 days ago

jambayk requested a review from

hariharans29 13 days ago

hariharans29 commented on 2026-03-19

Remove dead UseKleidiAI(HasZp) overload, rename UseKleidiAIBase

75c0b38d

jambayk requested a review from

hariharans29 13 days ago

hariharans29 approved these changes on 2026-03-19

jambayk merged d9bb7602 into main 12 days ago

jambayk deleted the jambayk/mnb-arm-32 branch 12 days ago

Reviewers

hariharans29

copilot-pull-request-reviewer

Assignees

No one assigned

Labels

None yet

Milestone

No milestone

onnxruntime Enable KleidiAI for asymmetric 4-bit MatMulNBits on ARM64 #27751 Merged

Enable KleidiAI for asymmetric 4-bit MatMulNBits on ARM64 #27751

onnxruntime
Enable KleidiAI for asymmetric 4-bit MatMulNBits on ARM64
#27751

Merged