Integration with ONNX rel-1.18.0

[titaiwangms]

Description

The PR adds CPU support by following release logics in https://github.com/onnx/onnx/wiki/Logistics-for-ONNX-Release-1.18.0. The goal is to do the minimal changes needed to ensure ONNXRUNTIME works fine with ONNX 1.18.0

Motivation and Context

Essentially, incoming ONNX 1.18.0 provides the following
(1) Introduce opset 23 (included in this PR)
(2) Support Attention, RMSNormalization, and RotaryEmbedding (NOT included in this PR)
(3) Support float4e2m1 (NOT included in this PR)

Remaining Issues

1. onnx.patch

• ONNXRUNTIME is using static functions (shape inference) from ONNX (ORT uses static shape inference functions in ONNX==1.18 #24558)
• GroupNormalization-18 is deprecated because its spec was wrong (GroupNormalization-18 is deprecated since ONNX==1.18.0 #24560)
• Contrib op registration api from ONNX: OpSchemaRegisterOnce is changed to explicit, and ONNXRUNTIME was leveraging it to do fluent-chaining style. (Update ORT to handle explicit OpSchemaRegisterOnce API in ONNX >= 1.18.0 for fluent chaining #24561)

2. Support float4e2m1 (Support FLOAT4E2M1 #24553)
3. Support Attention([Feature Request] Implement Attention-23 #24554), RMSNormalization([Feature Request] Implement RMSNormalization-23 #24555), and RotaryEmbedding([Feature Request] Implement RotaryEmbedding-23 #24556)
4. Disable QNN tests

[jywu-msft]

This is not targeted for OnnxRuntime 1.22 release, right?

[titaiwangms]

This is not targeted for OnnxRuntime 1.22 release, right?

No, it's not. ONNX 1.18.0 still on rc.

[Copilot]

Pull Request Overview

This PR integrates ONNX Runtime with ONNX rel‑1.18.0 by adding minimal changes required for CPU support and updating opset registrations consistent with the new release. Key changes include:

• Updating CPU control flow kernels (Scan, Loop, If) to support opset 23 for float4e2m1.
• Adjusting max supported opset values in core components and registering new opset entries in the operator tables.
• Refreshing documentation and dependency pointers to reflect the latest ONNX commit.

Reviewed Changes

Copilot reviewed 33 out of 40 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
onnxruntime/core/providers/cpu/controlflow/scan_9.cc	Updated kernel registration macro and added opset 23 kernel entry with a TODO for float4e2m1 support.
onnxruntime/core/providers/cpu/controlflow/loop.cc & if.cc	Similar updates in control flow kernels to support opset 23.
onnxruntime/core/providers/coreml/builders/impl/base_op_builder.h	Increased max supported opset from 21 to 23.
onnxruntime/core/optimizer/transpose_optimization/optimizer_api.h	Updated kMaxSupportedOpset from 22 to 23.
onnxruntime/core/optimizer/qdq_transformer/qdq_util.cc	Extended Q/DQ node version support to include opset 23.
onnxruntime/core/optimizer/layout_transformation/layout_transformation_potentially_added_ops.h	Added opset 23 entries for several operators.
graph.cc	Added an include for onnx/defs/parser.h as required by updated ONNX.
js/web/script/generate-webgl-operator-md.ts & js/web/docs/webgl-operators.md	Updated generated documentation references for operator definitions.
docs/OperatorKernels.md	Revised version ranges for several operators to include opset 23.
docs/How_To_Update_ONNX_Dev_Notes.md	Updated instructions to encourage using commit IDs for ONNX submodule updates.
cmake/deps.txt	Changed the ONNX dependency URL and commit hash to match the updated version.

Files not reviewed (7)

• cmake/external/onnx: Language not supported
• cmake/patches/onnx/onnx.patch: Language not supported
• cmake/vcpkg-ports/onnx/binskim.patch: Language not supported
• cmake/vcpkg-ports/onnx/fix-cmakelists.patch: Language not supported
• cmake/vcpkg-ports/onnx/fix-dependency-protobuf.patch: Language not supported
• cmake/vcpkg-ports/onnx/portfile.cmake: Language not supported
• cmake/vcpkg-ports/onnx/vcpkg.json: Language not supported

Comments suppressed due to low confidence (2)

onnxruntime/core/providers/cpu/controlflow/scan_9.cc:547

• Consider adding test coverage or tracking an issue for implementing float4e2m1 support in the opset 23 kernel registration to ensure the feature is validated once implemented.

// TODO(titaiwang): Add support for float4e2m1.

js/web/script/generate-webgl-operator-md.ts:82

• [nitpick] The updated script reference improves clarity by explicitly naming the generator; ensure similar explicit naming is used consistently in other documentation references.

def files via [generate-webgl-operator-md.ts](../script/generate-webgl-operator-md.ts).

[xadupre]

We should really simplify the patches. I assume everyone willing to link with ONNX should go through the sames patches. Looking at the list, it is long and hard to maintain.