[Justin Chu (justinchuby)]

Summary

• Generate the opset 28 API and expose standard opsets 24–28 at the package root. Keep unreleased opset 29 excluded.
• Refresh opset 25 to include Attention's sliding-window attributes.
• Add FLOAT6E2M3/FLOAT6E3M2 annotations and export all existing low-bit dtype annotations.
• Require onnx_ir 1.1.0, which introduced float6 support and transitively requires ml_dtypes 0.5.0 or later; align the CI ONNX IR pin.
• Preserve ONNX 1.17 compatibility by filtering schema type strings against installed ONNX enums and resolving annotation names through IR.
• Document the supported APIs, dtype requirements, and API generation command excluding opset 29.

Stack

Registered as a native GitHub stack in this order:

1. This PR: APIs, dtypes, documentation, and dependencies; targets main.
2. #3139: version-converter implementation, documentation, and regression coverage; targets this PR's branch.

This PR's diff contains no version-converter changes. The combined stack preserves the complete implementation from before the split.

CI follow-up

• Split SwiGLU coverage from the SpaceToDepth mode-attribute check. The pinned ONNX weekly build reports opset 28 but predates that attribute, so its availability is now checked against the actual installed schema. SwiGLU coverage still runs.
• Classify backend roundtrips with Optional graph inputs/outputs as unsupported by the scripting frontend, matching its existing annotation limitations. This does not add Optional-I/O frontend support or disable tests based on the presence of Optional operators alone.
• The Phi export schema-revision regression is fixed in the converter layer, Complete native ONNX opset conversion through 28 #3139. Inspected failed stable/nightly job logs contain no native ORT crash signatures; no ORT patch is justified by that evidence.

Validation

• Exact pinned CI ONNX weekly 1.23.0.dev20260831: complete backend roundtrip, API, and dtype suites 1104 passed, 760 skipped, no failures, with only the blanket Windows backend skip removed to exercise Linux-like coverage. Existing unsupported frontend/runtime cases remain skipped; the unavailable SpaceToDepth mode check skips independently.
• Current ONNX 1.24 API suite: 7 passed, including the SpaceToDepth mode check.
• Earlier independent bottom-layer integration across scripting/builders, IR schemas, tensors, API/dtype exports, and the unchanged converter suite: 295 passed, 1 skipped, 7 existing xfails.
• Every concrete IR dtype has a public annotation; numeric storage roundtrips include float6 and other low-bit types. API/dtype compatibility was also covered by earlier integration runs on current ONNX and minimum ONNX 1.17.
• lintrunner -a and git diff --check pass. Local ORT 1.31.1 is newer than the CI runtime pins; fresh hosted CI remains authoritative for those versions.

[codecov]

Codecov Report

❌ Patch coverage is 68.99563% with 71 lines in your changes missing coverage. Please review.
✅ Project coverage is 73.25%. Comparing base (682b59d) to head (1eb1149).
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
onnxscript/onnx_opset/_impl/opset28.py	51.69%	57 Missing ⚠️
onnxscript/onnx_opset_test.py	81.66%	10 Missing and 1 partial ⚠️
onnxscript/onnx_opset/_impl/opset25.py	62.50%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3138      +/-   ##
==========================================
- Coverage   73.28%   73.25%   -0.04%     
==========================================
  Files         267      269       +2     
  Lines       33214    33441     +227     
  Branches     3183     3192       +9     
==========================================
+ Hits        24342    24498     +156     
- Misses       7802     7874      +72     
+ Partials     1070     1069       -1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

[Copilot]

🟡 Changes recommended

The SpaceToDepth wrapper unconditionally forwards mode, causing even default DCR calls to fail with KeyError: 'mode' against the pinned ONNX schema, which lacks that attribute.

1 open issue

• Omit unsupported mode for SpaceToDepth default · New

What changed in this PR

Extends ONNXScript’s public APIs through opset 28 and completes low-bit dtype exports. This provides the API and dependency foundation for the separate version-converter changes in #3139.

Changes:

• Adds generated opset 28 APIs, exports opsets 24–28, and refreshes Attention’s sliding-window attributes.
• Adds float6 annotations, completes dtype exports, and updates ONNX IR requirements while preserving ONNX 1.17 compatibility.
• Expands API and dtype tests, adjusts backend coverage for unsupported Optional I/O, and documents availability and generation.

File	Description
pyproject.toml	Raises the ONNX IR minimum for float6 support.
opgen/​README.md	Documents generation excluding opset 29.
onnxscript/​onnx_types.py	Adds float6 annotations and resolves dtype names through IR.
onnxscript/​onnx_types_test.py	Tests dtype exports, annotations, and storage roundtrips.
onnxscript/​onnx_opset/​_impl/​opset28.py	Adds generated opset 28 operator wrappers.
onnxscript/​onnx_opset/​_impl/​opset25.py	Adds Attention sliding-window parameters and refreshes generated documentation.
onnxscript/​onnx_opset/​__init__.py	Registers and exports opset 28.
onnxscript/​onnx_opset_test.py	Tests public opsets, schema signatures, and new APIs.
onnxscript/​backend/​onnx_export_test.py	Skips scripting roundtrips with Optional graph I/O.
onnxscript/​_internal/​type_annotation.py	Filters schema dtype strings against installed ONNX enums.
onnxscript/​__init__.py	Exports recent opsets and remaining low-bit annotations.
noxfile.py	Aligns the CI ONNX IR pin with the new minimum.
docs/​api/​opsets.md	Documents public opsets, dtype exports, and requirements.

_{Review effort: Balanced}