Skip to content

Pin ORT binary to 1.23.0 to avoid KleidiAI conv memory regression in 1.24.x#1

Merged
biendltb merged 1 commit into
mainfrom
ort-1.23.0
Jul 11, 2026
Merged

Pin ORT binary to 1.23.0 to avoid KleidiAI conv memory regression in 1.24.x#1
biendltb merged 1 commit into
mainfrom
ort-1.23.0

Conversation

@biendltb

Copy link
Copy Markdown

ORT 1.24.x inflates peak memory 3-4x for fully-convolutional models on SME-capable ARM64 devices microsoft/onnxruntime#29538. The objectiveC/ bindings are byte-identical between upstream v1.23.0 and v1.24.2, so only the binary target URL and checksum change.

Revert this fork once an ORT release ships with microsoft/onnxruntime#28571 (expected 1.28) and an official SPM tag exists.

Related issue: masicai/flutter_onnxruntime/issues/66

…1.24.x

ORT 1.24.x inflates peak memory 3-4x for fully-convolutional models on
SME-capable ARM64 devices (microsoft/onnxruntime#29538). The objectivec/
bindings are byte-identical between upstream v1.23.0 and v1.24.2, so only
the binary target URL and checksum change.

Revert this fork once an ORT release ships with
microsoft/onnxruntime#28571 (expected 1.28) and an official SPM tag exists.
@biendltb
biendltb merged commit df20353 into main Jul 11, 2026
@biendltb
biendltb deleted the ort-1.23.0 branch July 11, 2026 13:05
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant