All pull requests

Pull requests

• [torchlib] Implement fmax, fmin and hypot

  #3169·Shawn (ShawnNotFound) opened Nov 6, 2026·

  ·

• Narrow multi-output match candidates to consumers of bound inputs

  #3168·Mayuri (mayuriphad) opened Nov 6, 2026·

  ·
  1

  1

• [Rewriter] Adopt missing ONNX Light optimization patterns

  #3167·Justin Chu (justinchuby) opened Nov 5, 2026Collaborator·

  ·
  27

• [torchlib] Add aten::upsample_linear1d.vec

  #3166·Kyunggeun Lee (kyunggeunlee) opened Nov 5, 2026·

  ·
  1

  1

• chore(deps): bump onnx-weekly from 1.23.0.dev20260831 to 1.24.0.dev20261102 in /requirements/ci

  dependenciespython
  #3164·dependabot[bot] opened Nov 5, 2026Bot·

  ·

• Guard mixed-precision skip RMS normalization fusion

  module: optimizer
  #3163·justinchuby with CopilotAI opened Nov 5, 2026·

  ·
  1

  4

  

• Use optimized reference ops in constant folding evaluator

  #3162·Santosh Todkar (todkarsant) opened Nov 5, 2026·3 tasks·

  ·
  1

  1

• Fix ConcatFromSequence lowering for opsets 11 and 12

  #3161·DiadkoShmek opened Nov 5, 2026·

  ·

• Add op test for torch.unique_consecutive

  #3160·Santosh Todkar (todkarsant) opened Nov 5, 2026·2 tasks·

  ·
  1

  2

• [rewriter] Skip FusedMatMul batch-transpose fusion on mismatched ranks

  #3159·Syed Mudassir (Syedmudassir1604) opened Nov 5, 2026·

  ·
  1

• fix: guard skip-rmsnorm fusion against mismatched dtypes (#3155)

  #3158·aanelluri opened Nov 4, 2026·

  ·
  1

• Keep the output Reshape when fusing MultiHeadAttention into a non-3D output

  #3157·Sankur Kundu (Shakhtar-Sankur) opened Nov 4, 2026·

  ·
  1

• Rewriter: fix cast_constant_of_shape for values outside the target range

  #3153·Lim Jooyoung (mmporong) opened Nov 3, 2026·

  ·

• [torchlib] Fix atan2 for signed zeros, infinities and NaN inputs

  #3152·Shawn (ShawnNotFound) opened Nov 3, 2026·

  ·
  1

• Document supported script syntax and expand compiler/eager parity tests

  #3150·G. Ramalingam (gramalingam) opened Nov 2, 2026Collaborator·

  ·
  6

• Isolate temporary evaluators and track eager graph function contexts

  #3149·G. Ramalingam (gramalingam) opened Nov 2, 2026Collaborator·

  ·
  7

• [torchlib] Keep float64 precision in log2 and log10

  #3148·Syed Mudassir (Syedmudassir1604) opened Nov 2, 2026·

  ·
  1

  1

• fix(rewriter): preserve PRelu data-input expansion

  #3147·DiadkoShmek opened Nov 2, 2026·

  ·

• Rewriter: guard string target and reference attributes in cast_constant_of_shape

  #3145·Hardik K (hardikkaurani) opened Nov 1, 2026Contributor·

  ·
  10

• Rewriter: don't fuse Min/Max into Clip when a bound has more dims than the input

  #3142·Shreya Boyane (Sonica-B) opened Oct 31, 2026Contributor·

  ·

• fix: preserve Conv parameters when fusing broadcast scalars

  #3141·DiadkoShmek opened Oct 31, 2026·

  ·
  1

• Complete native ONNX opset conversion through 28

  #3139·Justin Chu (justinchuby) opened Oct 30, 2026Collaborator·2/2·

  ·
  4

• Validate core export overrides and clarify public API contracts

  #3137·G. Ramalingam (gramalingam) opened Oct 29, 2026Collaborator·

  ·
  4

• chore(deps): bump onnxruntime from 1.25.1 to 1.31.1 in /requirements/ci

  dependenciespython
  #3135·dependabot[bot] opened Oct 29, 2026Bot·

  ·

• Fix ORT GQA mask contracts and FusedMatMul division boundaries

  #3134·G. Ramalingam (gramalingam) opened Oct 29, 2026Collaborator·

  ·
  5