All pull requests

Pull requests

• [Testcase Refactoring] Route FSDP test helpers through the selected device

  open sourcerelease notes: distributed (fsdp)
  #204408·danieldale2026 opened Nov 11, 2026Contributor·

  ·
  1

• Fix quantized::mul XNNPACK path for non-contiguous inputs

  bot-triagedmissing actionable issuemodule: cpuopen sourcerelease notes: quantizationtriaged
  #204402·gaurav-redhat opened Nov 11, 2026Contributor·1 task·

  ·
  1

  1

• [docs] Explain wrapper Tensor flatten/unflatten protocol

  open sourcetopic: not user facing
  #204401·fineverse opened Nov 11, 2026·

  ·
  1

  2

• [ROCm] Use top-left causal mask in CK memory-efficient attention

  module: rocmopen source
  #204400·mgehre-amd opened Nov 11, 2026Contributor·

  ·
  2

• [XPU] Enable FP8 scaled matmul tests

  ciflow/periodic-strictopen sourcerelease notes: python_frontend
  #204399·libohao1201 opened Nov 11, 2026Contributor·

  ·
  2

• [OpenReg] Support AOTInductor on the openreg device

  open sourcetopic: not user facing
  #204398·can-gaa-hou opened Nov 11, 2026Collaborator·

  ·
  1

• [Inductor] Support PrivateUse1 devices in the AOTI C++ wrapper codegen

  ciflow/inductorciflow/torchtitanmodule: inductoropen sourcerelease notes: inductor (aoti)
  #204397·can-gaa-hou opened Nov 11, 2026Collaborator·

  ·
  1

• [AOTI] Support PrivateUse1 devices in the model.so runtime

  module: aotinductormodule: PrivateUse1open sourcerelease notes: inductor (aoti)
  #204396·can-gaa-hou opened Nov 11, 2026Collaborator·

  ·
  2

• [quantization] Reject non-finite inputs in per-tensor min/max observers

  open sourcerelease notes: quantization
  #204395·tangyuan499-commits opened Nov 11, 2026·

  ·
  1

  1

• [quantization] Reject symmetric quint8 reduced-range HistogramObserver

  open sourcerelease notes: quantization
  #204394·tangyuan499-commits opened Nov 11, 2026·

  ·
  1

  1

• Register hash_tensor.out for XPU backend

  bot-triagedmissing actionable issueopen sourcetriaged
  #204393·msnliu opened Nov 11, 2026Contributor·

  ·
  4

• [CD] Add CUDA 13.5 Linux nightly binaries

  bot-triagedciflow/binariesciflow/dockerin progressopen sourcerelease notes: relengtriaged
  #204391·tinglvv opened Nov 11, 2026Collaborator·

  ·
  3

• test(inductor): make test_combo_kernels.py device-agnostic

  ciflow/inductormodule: inductoropen sourcetopic: not user facing
  #204390·li-lizhe opened Nov 11, 2026Contributor·

  ·
  2

• [XPU] Enable TestOptimRenewed in test_optim.py

  open sourcerelease notes: linalg_frontend
  #204387·NaOHCC opened Nov 11, 2026Contributor·

  ·
  1

• [testing] Fix largeTensorTest memory check on mismatched accelerators and make tests device-agnostic

  open sourcerelease notes: linalg_frontend
  #204386·NaOHCC opened Nov 11, 2026Contributor·

  ·
  1

• [no-ci] [cfg_compile] Keep kernel results unboxed until Python needs them, and add benchmarks

  #204385·bobrenjc93 opened Nov 11, 2026Contributor·

  ·
  2

• [cfg_compile] Add the Rust crate behind torch.cfg_compile: the SSA IR and a bytecode decoder

  #204384·bobrenjc93 opened Nov 11, 2026Contributor·

  ·
  2

• [cfg_compile] Call kernels through their C++ signature when it is a common one

  #204383·bobrenjc93 opened Nov 11, 2026Contributor·

  ·
  2

• [XPU][Inductor] Skip matmul output precision test on pre-Xe2 devices

  ciflow/inductormodule: inductormodule: xpuopen sourcetopic: not user facing
  #204382·lchen2331 opened Nov 11, 2026Contributor·

  ·
  1

  4

• [ROCm][FlexAttention] Tune gfx1250 prefill and preserve kernel options

  ciflow/inductorciflow/torchtitanmodule: inductormodule: rocmopen source
  #204381·nithinsubbiah opened Nov 11, 2026Contributor·

  ·
  4

• [no-ci] [cfg_compile] Let passes be written in Python

  #204379·bobrenjc93 opened Nov 11, 2026Contributor·

  ·
  2

• [no-ci] [cfg_compile] Profile blocks on real tensors and specialize them with a pass pipeline

  #204378·bobrenjc93 opened Nov 11, 2026Contributor·

  ·
  2

• [no-ci] [cfg_compile] Add the generic tier of the VM: a register interpreter over the SSA graph

  #204377·bobrenjc93 opened Nov 11, 2026Contributor·

  ·
  2

• [cfg_compile] Add the Rust frontend: Python bytecode to a control-flow graph in SSA form

  #204376·bobrenjc93 opened Nov 11, 2026Contributor·

  ·
  2

• [MKLDNN] Handle non-contiguous BatchNorm affine parameters

  module: cpumodule: mkldnnopen sourcerelease notes: cpu (x86)
  #204375·tangyuan499-commits opened Nov 11, 2026·

  ·
  1

  2