[webgpu] Add Matmul8bits Support #24546
sushraja-msft
merged
22 commits into
main
from
matmul8bits
[webgpu] Support nbits = 8 for matmulnbits
3bca9e8a
add 8bits support for non-dp4a path
eee685d3
add 8bits support for subgroup matrix path
48fc5c28
add nbits=8 support for MatMulNBitsWideTileProgram
502bd2cd
temporarily disable dp4 path for nbits = 8
3a5d892f
add zero points for MatMulNBitsBlockWiseProgram
450a5052
support any components of A for MatMulNBitsBlockWiseProgram
a5fa8ed3
remove all limitations of MatMulNBitsBlockWiseProgram
f3b2a4aa
Fix bugs in zero_points when nbits = 4
3a77a7ea
remove unused code
95aedbf4
Merge branch 'main' into matmul8bits
6077d228
enable tests for 8bits
f6c583bd
use flatten workgroup_idx
57457421
add todo
5e3e602b
sushanthr
commented
on 2025-04-28
sushanthr
commented
on 2025-04-28
sushanthr
commented
on 2025-04-28
fix the dp4 path overflow issue
186a79cc
address comments
98cc62d9
address comments
643b64aa
Merge branch 'main' into matmul8bits
5a396289
rename MatMulNBitsBlockWiseProgram to MatMulNBitsProgram
b078fd28
refactor ReadZeroPoint
1da9f83a
sushraja-msft
dismissed these changes
on 2025-05-02
sushraja-msft
commented
on 2025-05-05
address comments
1a378af7
fix the warning
07181a60
guschmue
approved these changes
on 2025-05-06
Reviewers
Assignees
No one assigned
Labels
ep:WebGPU
Milestone
No milestone
Login to write a write a comment.
Login via GitHub