Repository navigation

[executorch][native] Record dynamic-shape linears as chained bundled samples - #24549

Closed

pytorchbot wants to merge 128 commits into

gh/digantdesai/189/origfrom

gh/digantdesai/201/orig

Closed

[executorch][native] Record dynamic-shape linears as chained bundled samples#24549
pytorchbot wants to merge 128 commits into
gh/digantdesai/189/origfrom
gh/digantdesai/201/orig

Commits on Nov 3, 2026
- [ET-VK][conv2d] Use a 2x2x8 tile and tuned local sizes for pointwise convolution
  SS-JIA
  committed
  b52e504
- Qualcomm AI Engine Direct - Support QAT for concat observer (#23347)
  shewu-quic
  authored
  b6f80e3
- Qualcomm AI Engine Direct - Add composite lowering for aten.convolution_backward (#23610)
  
  authored
  18e0cc0
- Arm backend: Add CONST and use tosa dialect op (#24287)
  AdrianLundell
  authored
  274a774
- Arm backend: Collect PMU counters in EthosUExecutionStats (#24432)
  Erik-Lundell
  authored
  0250ab2
- Arm backend: Fix metadata provenance for DecomposeTopK (#24497)
  wwwind
  authored
  a489dca
- Arm backend: Group adaptive average pooling output windows (#24498)
  AdrianLundell
  authored
  a3528d9
- Reinstall pip from PyPI in the native wheel pre-build script (#24461)
  JakeStevens
  authored
  6d903f4
- Keep max_pool2d and avg_pool2d test inputs within the exported shape bounds (#24483)
  digantdesai
  authored
  2799a79
- Arm backend: Add FP8 to automatic list of generated ops, VGF. (#24496)
  wwwind
  authored
  e655cb8
- Separate ESP32 CI runs by commit (#24472)
  rascani
  authored
  a057bb4
- Run Espressif operator-library cases in optimized S3 CI (#24313)
  rascani
  authored
  7da27ca
- XNNPACK: read inputs with a non-default dim order in their memory layout (#24081)
  msluszniak
  authored
  249237c
- Keep int64 precision in portable div trunc mode and fmod (#24170)
  usmanmateen
  authored
  186ca41
- Fix NXP MCUXpresso CIFAR link failure (#24510)
  JacobSzwejbka
  authored
  ee4b85b
- Fix export of zero-stride constants and inputs, and of constants that delegates share (#24404)
  shoumikhin
  authored
  24ad53e
- Fix wrong dtypes and values in Core ML conversion (#24424)
  shoumikhin
  authored
  3bc5222
- Enable Vulkan vision runners on Android (#24503) (#24503)
  JakeStevens
  authored
  a8f64b0
- Move backend-neutral Cadence passes to backends/transforms (#24469)
  mcremon-meta
  authored
  a1772a6
- Python runtime: stop a backend's log from deadlocking 2 threads (#24393)
  shoumikhin
  authored
  f1240a2
- Support Whisper positional indexing (#23903) (#23903)
  JakeStevens
  authored
  d55e079
- CUDA backend: lower rotary embeddings written with complex numbers (#24403)
  shoumikhin
  authored
  c68298d
- XNNPACK: fix wrong outputs when BatchNorm constants share a name, a bf16 matmul load crash, and skipped bmm tests (#24504)
  shoumikhin
  authored
  bd0fcde
- Remove dead CUDA 11 select arm
  r-barnes
  authored
  97e9fbb
- Arm backend: Fix provenance for rSub.Scalar (#24500)
  wwwind
  authored
  27de6db
- NXP backend: Enable MobileNetV3 model (#24307)
  roman-janik-nxp
  authored
  892e504
- Remove the unused iDMA ExecuTorch ops
  mcremon-meta
  authored
  33de505
- CI: stop the torch-free bindings check from matching load addresses (#24470)
  shoumikhin
  authored
  5badf54
- Mark the 3-D convolution cases that fail inside a backend as expected failures (#24459)
  shoumikhin
  authored
  6c3242e
- [ET-VK][benchmark] Rename migration shader targets to prototype_shaders
  
  SS-JIA
  authored and
  pytorchbot
  committed
  df0cc0f
- [ET-VK][benchmark] Model the q8csw weight-only fallback in references
  
  SS-JIA
  authored and
  pytorchbot
  committed
  3411581
- [ET-VK][benchmark] Print shader bindings as a string in JSON summaries
  
  SS-JIA
  authored and
  pytorchbot
  committed
  5cc1f12
- [ET-VK][benchmark] Add sustained replay for hardware counter capture
  
  SS-JIA
  authored and
  pytorchbot
  committed
  f9dcc9f
- [ET-VK][ci] Run benchmark_v2 CI cases instead of etvk_bench in Skycastle
  
  SS-JIA
  authored and
  pytorchbot
  committed
  2bb5843
- [ET-VK][ci] Build benchmark_v2 with CMake and run it in GitHub CI
  
  SS-JIA
  authored and
  pytorchbot
  committed
  3dfe38d
- [ET-VK][benchmark] Port remaining legacy coverage to benchmark_v2
  
  SS-JIA
  authored and
  pytorchbot
  committed
  e1d5a50
- [ET-VK][benchmark] Remove the legacy operator benchmark
  
  SS-JIA
  authored and
  pytorchbot
  committed
  07e8b1e
- exir and MLX: fix folded-output clones, shared constants, boolean masks, int64 scatter_add, lowering time and memory, and #24404 follow-ups (#24481)
  shoumikhin
  authored
  79feb10
Commits on Nov 4, 2026
Commits on Nov 5, 2026

Commits

Commits on Nov 3, 2026

Commits on Nov 4, 2026

Commits on Nov 5, 2026