All pull requests

Pull requests

• Accept empty tensor bindings in CPUBackend execute (#24195)

  CLA Signedmeta-exported
  #24195·JakeStevens opened Oct 27, 2026Contributor·

  ·
  3

• Route _neg_view_copy to ET neg kernel (#24194)

  CLA Signedmeta-exported
  #24194·JakeStevens opened Oct 27, 2026Contributor·

  ·
  3

• Wire batch-norm fusion into CpuBackend.preprocess (#24193)

  CLA Signedmeta-exported
  #24193·JakeStevens opened Oct 27, 2026Contributor·

  ·
  3

• Add shared linear-batch-norm fusion pass (#24192)

  CLA Signedmeta-exported
  #24192·JakeStevens opened Oct 27, 2026Contributor·

  ·
  3

• Support integer and boolean CPU tensors (#24191)

  CLA Signedmeta-exported
  #24191·JakeStevens opened Oct 27, 2026Contributor·

  ·
  3

• [ET-VK][gemma4] Add ETDump profiling

  CLA Signed
  #24189·SS-JIA opened Oct 27, 2026Contributor·3/3·

  ·
  1

• [ET-VK][SDPA] Preserve shared KV cache semantics

  CLA Signed
  #24188·SS-JIA opened Oct 27, 2026Contributor·2/3·

  ·
  1

• [ET-VK][embedding] Support per-row Q4 scales

  CLA Signed
  #24187·SS-JIA opened Oct 27, 2026Contributor·1/3·

  ·
  2

• Arm backend: Reuse frame features in QAT example

  ciflow/trunkCLA Signedmodule: armpartner: armrelease notes: none
  #24186·oscarandersson8218 opened Oct 27, 2026Collaborator·

  ·
  1

• [cuda] Decompose depthwise conv2d into pointwise Triton kernels

  ciflow/cudaCLA Signedrelease notes: backends
  #24185·shoumikhin opened Oct 27, 2026Contributor·

  ·
  5

• Update arm .wiki information

  CLA Signedpartner: armrelease notes: none
  #24184·AdrianLundell opened Oct 27, 2026Collaborator·

  ·
  4

• [XNNPACK] Keep NHWC across scalar and low-rank broadcast operands

  CLA Signedcommunity: contributionrelease notes: xnnpack
  #24183·borys3kk opened Oct 27, 2026Contributor·

  ·
  2

• Fix two crashes caused by constant_prop_pass

  CLA Signedrelease notes: exir
  #24182·shoumikhin opened Oct 27, 2026Contributor·

  ·
  2

• Serialize only a view constant's own bytes in XNNPACK and Vulkan

  CLA Signedmodule: vulkanrelease notes: xnnpack
  #24181·shoumikhin opened Oct 27, 2026Contributor·

  ·
  2

• Keep the FACTO pow exponent representable in float32

  CLA Signedmeta-exported
  #24179·aliafzal opened Oct 27, 2026Contributor·

  ·
  3

• Treat a missing rounding mode in div_out_mode as true division

  CLA Signedmeta-exported
  #24178·aliafzal opened Oct 27, 2026Contributor·

  ·
  3

• Skip ReplacePowWithMulPass when pow promotes the input dtype

  CLA Signedmeta-exported
  #24177·aliafzal opened Oct 27, 2026Contributor·

  ·
  3

• Move positional alpha to kwargs in ReplaceScalarWithTensorArgPass

  CLA Signedmeta-exported
  #24176·aliafzal opened Oct 27, 2026Contributor·

  ·
  3

• [ET][native] Verify bundled LLM samples through the runner

  CLA Signedmeta-exported
  #24175·digantdesai opened Oct 27, 2026Contributor·

  ·
  2

• [ET][native] Bundle LLM prefill and decode samples from export_hf_llm

  CLA Signedmeta-exported
  #24174·digantdesai opened Oct 27, 2026Contributor·

  ·
  2

• [ET][native] Read bundled samples in C++

  CLA Signedmeta-exported
  #24173·digantdesai opened Oct 27, 2026Contributor·

  ·
  2

• [ET][native] Make the LLM runner's next-token choice pluggable

  CLA Signedmeta-exported
  #24172·digantdesai opened Oct 27, 2026Contributor·

  ·
  2

• [ET][native] Bundle sample method calls in .ptn packages

  CLA Signedmeta-exported
  #24171·digantdesai opened Oct 27, 2026Contributor·

  ·
  2

• Keep int64 precision in portable div trunc mode and fmod

  CLA Signedcommunity: contribution
  #24170·usmanmateen opened Oct 27, 2026Contributor·

  ·
  1

  2

• Keep float scalars exact in float64 ops in ScalarToTensorPass

  CLA Signedcommunity: contribution
  #24168·usmanmateen opened Oct 27, 2026Contributor·

  ·
  1

  2