All pull requests

Pull requests

• Fix unsafe Slice/MatMul reordering with broadcasted operands

  #367·HowieYin opened Oct 22, 2026Contributor·

  ·

• Fix nonzero Shape.start offset in Slice folding

  #365·sean0407 opened Oct 21, 2026·

  ·
  1

• fix: skip vector-weight fuse_qkv patterns

  #364·sean0407 opened Oct 21, 2026·

  ·

• Prevent fusion of opset 10 Pad with nonzero value

  #363·sean0407 opened Oct 21, 2026·

  ·
  1

• Fix common-subexpression elimination across domains

  #362·sean0407 opened Oct 21, 2026·

  ·
  1

• Add eliminate_initializer_from_input pass

  #325·take-cheeze opened Jul 29, 2026Member·

  ·
  9

• Add optional graph-modification report to optimize()

  #324·take-cheeze opened Jul 29, 2026Member·

  ·
  6

• Add fuse_attention pass for torch.onnx SDPA graphs

  #321·take-cheeze opened Jul 26, 2026Member·

  ·
  10

• Add fuse_gelu pass to fuse erf-based GELU into a Gelu op

  #320·take-cheeze opened Jul 26, 2026Member·

  ·

• Add C++ IR (Graph) entry point to the optimizer

  #319·take-cheeze opened Jul 26, 2026Member·

  ·
  1

• Fix fuse_pad_into_pool folding zero-padding into MaxPool

  bug
  #318·take-cheeze opened Jul 24, 2026Member·

  ·
  1

  1

• Add tests for ONNX function preservation in optimizer

  #317·take-cheeze opened Jul 24, 2026Member·

  ·
  1

  1

• Support fuse bn into ConvTranspose.

  #316·take-cheeze opened Jul 23, 2026Member·

  ·
  1

• Fix split_predict generating invalid ONNX models with missing elem_type

  bug
  #244·justinchuby with CopilotAI opened Dec 31, 2025·

  ·
  1

  4

  

• Support fuse bn into ConvTranspose.

  bug
  #106·YuchiWen opened Nov 22, 2022·

  ·
  6

• Add a new fusion pass with pattern of constant and reshape

  bug
  #62·jiangjiajun opened Feb 23, 2022·

  ·
  4

• support fusing non constant bias into conv

  bug
  #37·daquexian opened Apr 4, 2021Member·

  ·
  8