All pull requests

Pull requests

• Fix nondeterministic ONNX node naming

  #1084·ljfitz opened Oct 21, 2026Collaborator·

  ·

• [release_rai_1_8] Transpose/requantize name handling and fp16 Resize/Reduce->Conv fixes

  #1075·rachgupt-amd opened Oct 12, 2026·

  ·

• Keep ONNX dim_params index-aligned and preserve metadata_props

  #1072·kartikpodugu-amd opened Oct 7, 2026·2 tasks·

  ·

• Decompose Gridsample with volumetric input.

  #1059·SamNour opened Sep 29, 2026·

  ·
  1

  

• apply canonicalization again

  #1037·ayanshul-amd opened Sep 16, 2026Collaborator·

  ·

• [AIESW-38387] testing hardsigmoid

  #1036·chaitanyakamarapu opened Sep 16, 2026Collaborator·

  ·

• Allow narrower argmax result dtypes

  #1031·ljfitz opened Sep 15, 2026Collaborator·

  ·
  1

• Backport xiao.uplft gather

  #1021·xiaoh-amd opened Sep 3, 2026·

  ·

• Recompose nested Where clamp idiom into onnx.Clip

  #1018·ljfitz opened Sep 3, 2026Collaborator·

  ·
  9

• Add conv kernel width support condition

  #985·pragyam32 opened Aug 12, 2026·

  ·

• Adjust to arith const argument order change

  #979·jorickert opened Aug 10, 2026·

  ·

• add com.microsoft.GatherBlockQuantized decomposition

  #967·gerion-amd opened Aug 5, 2026·

  ·

• Bump prep4

  #964·jorickert opened Aug 4, 2026·

  ·

• fix: avoid integer add while creating attention mask

  #928·ge28boj opened Jul 20, 2026·

  ·

• Add slice canonicalization patterns

  #927·tr00 opened Jul 20, 2026·

  ·
  16

• [testing] Revert "Fix semantic gaps of microsoft GQA decomposition"

  #920·ge28boj opened Jul 16, 2026·

  ·

• Sameek/doc

  #893·sambanerxilinxamd opened Jul 7, 2026·

  ·

• Rewrite AvgPool with ceil_mode=1 to ceil_mode=0 + end padding

  #879·tr00 opened Jul 1, 2026·

  ·
  4

• Skip Mul fusion for HSigmoid op

  #860·pragyam32 opened Jun 25, 2026·

  ·

• Q-DQ constant folding with diff quant params

  #806·ilango100 opened Jun 1, 2026·

  ·
  5

• changes to convtranspose decomposition

  #794·chaitanyakamarapu opened May 28, 2026Collaborator·

  ·

• Add combined RedundantOpReductionPass mirroring xcompiler exec() for QDQ models

  #784·rachgupt-amd opened May 26, 2026·

  ·

• fix: add golden guards to MergeStridedSliceConcatConv and MergeSliceCononcat passes

  #781·rachgupt-amd opened May 24, 2026·

  ·

• DequantizeLinear: infer Y’s shape from X

  #749·SamNour opened May 8, 2026·

  ·
  1

• Temporarly keep layernorm decomposed

  #725·chaitanyakamarapu opened Apr 29, 2026Collaborator·

  ·