onnxruntime
Route fp16 HQNBIT_CompInt8 (4-bit and 8-bit) through fp32 MLAS path in MatMulNBits
#27820

Merged

Route fp16 HQNBIT_CompInt8 (4-bit and 8-bit) through fp32 MLAS path in MatMulNBits #27820

jambayk merged 5 commits into main from jambayk/mnb-4-16

Route fp16 HQNBIT_CompInt8 through fp32 MLAS path for 4-bit and 8-bit

c0c51bed

jambayk requested a review from

copilot-pull-request-reviewer 90 days ago

github-actions commented on 2026-03-24

copilot-pull-request-reviewer commented on 2026-03-24

Remove dead HQ4BitGemm_CompInt8 and HQ8BitGemm_CompInt8 MLAS code

a217afb1

lint

0931a5f2

jambayk requested a review from

copilot-pull-request-reviewer 90 days ago

copilot-pull-request-reviewer commented on 2026-03-24

Fix HQNBIT_CompInt8 PrePack bugs for 4-bit and 8-bit

316b13fa

jambayk requested a review from

copilot-pull-request-reviewer 89 days ago

copilot-pull-request-reviewer commented on 2026-03-24

jambayk marked this pull request as ready for review 89 days ago

jambayk requested a review from

hariharans29 89 days ago

jambayk requested a review from

vraspar 89 days ago

jambayk enabled auto-merge (squash) 89 days ago

vraspar commented on 2026-03-24

vraspar dismissed these changes on 2026-03-24

hariharans29 commented on 2026-03-24

hariharans29 dismissed these changes on 2026-03-24

jambayk dismissed their stale review via 23094525 89 days ago

jambayk force pushed from fc606f56 to 23094525 89 days ago

vraspar dismissed these changes on 2026-03-24

Address review: ORT_ENFORCE for scales, move SQNBIT check to GetCompu…

138318a0

jambayk dismissed their stale review via 138318a0 89 days ago

jambayk force pushed from 23094525 to 138318a0 89 days ago

vraspar approved these changes on 2026-03-25

jambayk merged 36242c6c into main 88 days ago

jambayk deleted the jambayk/mnb-4-16 branch 88 days ago

Reviewers

vraspar

hariharans29

github-actions

copilot-pull-request-reviewer

Assignees

No one assigned

Labels

None yet

Milestone

No milestone

onnxruntime Route fp16 HQNBIT_CompInt8 (4-bit and 8-bit) through fp32 MLAS path in MatMulNBits #27820 Merged

Route fp16 HQNBIT_CompInt8 (4-bit and 8-bit) through fp32 MLAS path in MatMulNBits #27820

onnxruntime
Route fp16 HQNBIT_CompInt8 (4-bit and 8-bit) through fp32 MLAS path in MatMulNBits
#27820

Merged