onnxruntime
Add fp16 support for 8-bit MatMulNBits on ARM64 and fix pre-existing bugs
#27692
Merged
Merged
Add fp16 support for 8-bit MatMulNBits on ARM64 and fix pre-existing bugs #27692
jambayk
merged
12 commits into
main
from
jambayk/mnb-arm-16
Add HQNBIT_CompFp16 support for 8-bit MatMulNBits on ARM64 NEON
93529bc8
acc 4 support for HQ
566df8df
fix bias bug, compint8 support for 8 bits
cd67587a
remove MLAS_TARGET_AMD64_IX86 guard on QuantBDataWorkspace
128908ae
fix scale packing
8415246a
only pack for 8 bit
6332d94a
copilot-pull-request-reviewer
commented
on 2026-03-17
jambayk
commented
on 2026-03-17
address reviews
218a6f9b
more reviews
ce7d5335
fix DequantB8Bit reference in dequant test
d4064093
Fix Float16_8b_ARM_CompFp16 SIGTRAP in Debug builds
886971a8
hariharans29
commented
on 2026-03-17
Add fp16 CompInt8 8-bit tests and improve N/K/BlockSize coverage
8359fe0b
hariharans29
dismissed these changes
on 2026-03-17
Increase fp16 8-bit test tolerances for large-K cases
e0a1834d
hariharans29
approved these changes
on 2026-03-17
Reviewers
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub