All pull requests

Pull requests

• [torchlib] Fix aten::any returning True for an empty reduction

  #3208·Mohammad Sohail Shaikh (sohailshk) opened Nov 13, 2026·

  ·

• [torchlib] Promote mixed dtypes in aten::pow.Tensor_Tensor

  #3207·Mohammad Sohail Shaikh (sohailshk) opened Nov 13, 2026·

  ·

• Add tool to generate rewrite rules from a model subgraph (#1814)

  #3206·Bhargav Kumar Nath (BhargavKumarNath) opened Nov 13, 2026·

  ·
  1

  1

• Fix consecutive cast elimination for inexact FLOAT intermediates

  #3205·itsrayeedscodespace opened Nov 12, 2026·

  ·
  1

• Rewriter: don't turn Flatten into Reshape when a dimension is zero

  #3203·dfedoryshchev opened Nov 12, 2026Contributor·

  ·

• Authenticate the package feed before official Python builds

  #3202·Justin Chu (justinchuby) opened Nov 12, 2026Collaborator·

  ·
  1

• [torchlib] Implement adaptive_avg_pool1d/2d/3d for any output size and dynamic input sizes

  module: torchlib
  #3201·Qiming Bao (14H034160212) opened Nov 12, 2026·

  ·
  1

• [torchlib] Use scalar CumSum axis and support empty repeats in aten_repeat_interleave_Tensor

  #3199·小哥不太逍遥 (xgbtxy) opened Nov 11, 2026·

  ·
  1

• Preserve double inputs in exported infinity checks

  #3198·josepha_mayo (josepha-mayo) opened Nov 11, 2026·

  ·
  1

• Fix first-run detection for unique_consecutive exports

  #3197·josepha_mayo (josepha-mayo) opened Nov 11, 2026·

  ·
  1

• [torchlib] Cast to dtype before Softmax/LogSoftmax in softmax and log_softmax

  #3196·Qiming Bao (14H034160212) opened Nov 11, 2026·

  ·

• Preserve overridable initializer defaults in rewrite rules

  #3195·Justin Chu (justinchuby) opened Nov 10, 2026Collaborator·

  ·
  1

• [Rewriter] Support default attribute matching and validate attribute safety in ScatterAllDynamic

  #3193·Hardik K (hardikkaurani) opened Nov 10, 2026Contributor·

  ·
  13

• Preserve unresolved function attributes in rewrite rules

  #3191·Justin Chu (justinchuby) opened Nov 10, 2026Collaborator·

  ·
  1

• Fix BART attention key layouts and shape-derived scaling

  #3190·Justin Chu (justinchuby) opened Nov 10, 2026Collaborator·

  ·
  1

• Preserve initializer semantics during constant folding

  #3189·Justin Chu (justinchuby) opened Nov 10, 2026Collaborator·

  ·
  2

• [Rewriter] Expose RewritePass verbosity and document diagnostics

  #3188·Justin Chu (justinchuby) opened Nov 10, 2026Collaborator·

  ·
  1

  1

• Fix Reshape shape materialization proof and opset boundaries

  #3187·Justin Chu (justinchuby) opened Nov 10, 2026Collaborator·

  ·
  1

• Fix attention and partial rotary fusion correctness

  #3186·Justin Chu (justinchuby) opened Nov 10, 2026Collaborator·

  ·
  3

  1

• Replace default ORT GroupNorm fusion with ONNX GroupNormalization

  module: optimizer
  #3183·justinchuby with CopilotAI opened Nov 10, 2026·

  ·
  1

  8

  

• [torchlib] Return 0 for NaN inputs in aten::heaviside

  #3180·Qiming Bao (14H034160212) opened Nov 10, 2026·

  ·

• [torchlib] Support bool inputs in aten::max.dim and aten::min.dim

  #3179·Qiming Bao (14H034160212) opened Nov 10, 2026·

  ·

• [torchlib] Accept integer inputs in ceil/floor/round/trunc and isnan/isfinite

  #3178·Qiming Bao (14H034160212) opened Nov 10, 2026·

  ·

• [torchlib] Add aten::_fused_rms_norm

  #3177·0z5a (0z5a) opened Nov 9, 2026·

  ·
  1

  1

• [torchlib] Keep precision of expm1 and log1p near zero

  #3175·0z5a (0z5a) opened Nov 9, 2026·

  ·
  1