Sync with Microsoft ONNX Runtime - 21082026 - #1265

Open

ai-fw-intg wants to merge 20 commits into

ovep-developfrom

sync_msft_21082026

Commits on Aug 19, 2026

[WebGPU] Enable GeluFusion and BiasGeluFusion for the WebGPU EP (#32053)

fanchenkong1
and
Copilot
authored
10ea381
[CUDA] upgrade cutlass 4.7 and cudnn-frontend 1.27 (#32111)
tianleiwu
authored
fccfa10
Use commit timestamps for plugin EP dev versions (#32095)

edgchen1
and
Copilot
authored
84e87b4
[Build] Update cuda plugin linux aarch64 parallel to 8 (#32165)
tianleiwu
authored
00b6cdb
Fix #31573, prevent ARM64 SymmQgemm int16 overflow (#32057)
sylvesterkaczmarek
authored
f73b9ef

Commits on Aug 20, 2026

Merge branch 'microsoft:main' into master
hdharpure9922
authored
cfd731d
Avoid overflow in CPU TensorScatter indices (#32012)
apsonawane
authored
f703c52
Canonicalize external data locations (#32135)
apsonawane
authored
d3b86e7
Handle empty initializer axis scaling (#32138)
apsonawane
authored
597e4a6
Skip overridable initializer fusion (#32143)
apsonawane
authored
8805b39
Reject scalar Normalizer inputs (#32161)
apsonawane
authored
c6a81e4
Validate Conv bias size (#32160)
apsonawane
authored
d9872d8
Validate empty reduction axes (#32156)
apsonawane
authored
113375e
Validate QEmbed segment inputs (#32144)
apsonawane
authored
924b4d1
Validate ScatterND index depth (#32034)
apsonawane
authored
d63396d
Retain Python async run resources (#32041)
apsonawane
authored
1cb4307
Merge branch 'microsoft:main' into master
sfatimar
authored
f9a389b
Validate per-element split sizes on the input-tensor path to prevent OOB read (#29461)

chilo-ms
and
Copilot
authored
abc74dc
Merge remote-tracking branch 'upstream/main'
AIFrameworksIntegration
committed
d199465
Merge remote-tracking branch 'origin/master' into sync_msft_21082026
AIFrameworksIntegration
committed
4937f05