ReleaseMicrosoftMicrosoftpublished Oct 16, 2025seen Jun 26

microsoft/onnxscript v0.5.4

microsoft/onnxscript

Open original ↗

Captured source

source ↗
published Oct 16, 2025seen Jun 26captured 6dhttp 200method plain

v0.5.4

Repository: microsoft/onnxscript

Tag: v0.5.4

Published: 2025-10-16T16:52:30Z

Prerelease: no

Release notes:

What's Changed

Optimizer and rewriter

  • Fix constant in constant folding by @titaiwangms in https://github.com/microsoft/onnxscript/pull/2622
  • Create helper for comparing semantic equivalence of shapes by @justinchuby in https://github.com/microsoft/onnxscript/pull/2620
  • Fix GQA fusion to produce present key/value by @justinchuby in https://github.com/microsoft/onnxscript/pull/2634

Torch Lib

  • Separated implementation of aten::scatter overloads by @linshokaku in https://github.com/microsoft/onnxscript/pull/2605
  • Enhanced type annotations and simplified implementation of scatter.value by @linshokaku in https://github.com/microsoft/onnxscript/pull/2612
  • support for scalar args to aten::scatter by @linshokaku in https://github.com/microsoft/onnxscript/pull/2613
  • [torchlib] Implement aten_bilinear function using Einsum by @Copilot in https://github.com/microsoft/onnxscript/pull/2574
  • Simplify aten_unbind when shape is static by @justinchuby in https://github.com/microsoft/onnxscript/pull/2597
  • Consolidate overloads in torchlib by @justinchuby in https://github.com/microsoft/onnxscript/pull/2604
  • [torchlib] Fix implementations for bitwise_* overloads by @justinchuby in https://github.com/microsoft/onnxscript/pull/2618
  • [torchlib] Deprecate Rank and IsScalar by @justinchuby in https://github.com/microsoft/onnxscript/pull/2624
  • [torchlib] Fix operator add by @justinchuby in https://github.com/microsoft/onnxscript/pull/2630
  • Remove redundant registration of operator::add and fix sub.Scalar by @justinchuby in https://github.com/microsoft/onnxscript/pull/2631

Other Changes

  • Update torch api error message to include value names by @justinchuby in https://github.com/microsoft/onnxscript/pull/2599
  • Remove beartype by @justinchuby in https://github.com/microsoft/onnxscript/pull/2603
  • Allow opset_version to be set explicitly when exporting by @NoRaincheck in https://github.com/microsoft/onnxscript/pull/2615
  • Merge shapes only in identity op and nodel-level shape inference by @titaiwangms in https://github.com/microsoft/onnxscript/pull/2623

Full Changelog: https://github.com/microsoft/onnxscript/compare/v0.5.3...v0.5.4