All pull requests

Pull requests

• Allow fixed-qparam observers to accept multi-element qparams

  CLA Signed
  #4684·SiavashShams opened Aug 3, 2026·

  ·
  3

• Refactor da8w4 test to use shared model architecture

  CLA Signed
  #4681·abdulahi-banji opened Aug 2, 2026·

  ·
  3

• make tensor subclass copy_ error report exactly which metadata mismatched

  CLA Signed
  #4664·sainikhiljuluri opened Jul 30, 2026·

  ·
  1

  4

• [test] Fix dead device comparison that disabled the SM80 bfloat16 skip in test/integration/test_integration.py

  ciflow/xpuCLA Signedmodule: not user facingxpu
  #4660·karol-brejna-i opened Jul 30, 2026·

  ·
  6

• Add core support for static Range Quantization (SRQ) and sub-byte embedding optimizations

  CLA Signed
  #4659·pculliton opened Jul 30, 2026·

  ·
  5

• [xpu][test][integration] Enable test_export_float8 on XPU

  ciflow/xpuCLA Signedmodule: not user facingxpu
  #4658·karol-brejna-i opened Jul 30, 2026·

  ·
  4

• Mask scale loads in blockwise fp8 training GEMMs

  CLA Signed
  #4648·guptaishaan opened Jul 29, 2026·

  ·
  1

  1

• [torchao] Add torch-free linear kernel target for ExecuTorch consumers

  CLA Signed
  #4645·digantdesai opened Jul 29, 2026Contributor·

  ·
  1

• [xpu][test][dtypes] Finish XPU enablement of test_affine_quantized_float

  ciflow/xpuCLA Signedmodule: not user facingxpu
  #4637·karol-brejna-i opened Jul 29, 2026·

  ·
  7

• [xpu][test][float8] Enable test_float8_utils on XPU

  ciflow/xpuCLA Signedmodule: not user facingxpu
  #4636·karol-brejna-i opened Jul 29, 2026·

  ·
  7

• [xpu][test] Enable int8 mixed-precision training tests (test/prototype/test_quantized_training.py) and register scaled_int8_mm for XPU

  ciflow/xpuCLA Signedmodule: trainingxpu
  #4630·karol-brejna-i opened Jul 28, 2026·

  ·
  4

• Make the QAT tests to be device agnostic

  ciflow/xpuCLA Signedmodule: not user facingxpu
  #4629·karol-brejna-i opened Jul 28, 2026·

  ·
  28

• [xpu][test] Use device-agnostic device list in test_quantized_training

  ciflow/xpuCLA Signedmodule: not user facingxpu
  #4628·karol-brejna-i opened Jul 28, 2026·

  ·
  7

• [mxfp8] fix dim1 scale store for tiles which do not divide the shape

  CLA Signed
  #4627·truong-v opened Jul 28, 2026·

  ·
  3

• Drop the zero-row copy in permute_and_pad

  CLA Signed
  #4623·indianspeedster opened Jul 27, 2026·

  ·
  2

• Forward weight dtype to gemlite helper in UIntxBitPackedTensor

  CLA Signed
  #4622·guptaishaan opened Jul 27, 2026·

  ·
  1

  1

• extract FP8 inference compute kernels into torch.library ops (#4599)

  CLA Signed
  #4619·TangSiLiYang opened Jul 27, 2026·

  ·
  1

• [Bug Fix] Inverted bias flag in swap_conv2d_1x1_to_linear

  CLA Signedmodule: inference
  #4618·gaurav-redhat opened Jul 27, 2026·2 tasks done·

  ·
  9

• fix(autocast):fix the bug when backward scaling is ScalingType.DISABLED

  CLA Signed
  #4617·TangSiLiYang opened Jul 27, 2026·

  ·
  4

• fix(kernel): decouple int_scaled_matmul and int_matmul op schemas from triton (#4602)

  CLA Signed
  #4612·GiGiKoneti opened Jul 25, 2026Contributor·

  ·
  1

  1

• docs/types(pt2e): annotate and document prepare, fake_quantize, and reference_representation_rewrite (#4541)

  CLA Signed
  #4611·GiGiKoneti opened Jul 25, 2026Contributor·

  ·
  1

• docs/types(pt2e): annotate and document utils and qat_utils (#4541)

  CLA Signed
  #4610·GiGiKoneti opened Jul 25, 2026Contributor·

  ·
  1

• docs/types(pt2e): annotate and document graph_utils and export_utils (#4541)

  CLA Signed
  #4609·GiGiKoneti opened Jul 25, 2026Contributor·

  ·
  1

• Support GPTQ synchronization on accelerator devices

  #4608·tarun-bandi opened Jul 25, 2026·

  ·
  1

  2

• fix(autocast): resolve autocast state from the input's device type (#4601)

  CLA Signed
  #4603·Anai-Guo opened Jul 24, 2026Contributor·

  ·
  1

  3