All pull requests

Pull requests

• Give the Windows unit test jobs 180 minutes instead of 120

  CLA Signedtopic: not user facing
  #24666·shoumikhin opened Nov 5, 2026Contributor·

  ·
  5

• Arm backend: Update IFNetTwoBlock lowering for explicit TOSA operands

  ciflow/trunkCLA Signedpartner: armrelease notes: none
  #24665·gggekov opened Nov 5, 2026Collaborator·

  ·
  1

• Samsung Exynos AI LiteCore- Samsung gemma3 llama quant

  CLA Signedmodule: samsungpartner: samsung
  #24660·Jiseong-oh opened Nov 5, 2026Collaborator·

  ·
  2

• Samsung Exynos AI LiteCore - test annoate qparam

  CLA Signedmodule: samsungpartner: samsung
  #24659·Jiseong-oh opened Nov 5, 2026Collaborator·

  ·
  3

• Fix addmm lowering when beta or alpha is not 1, or the bias is not 1-D

  CLA Signedrelease notes: xnnpack
  #24658·shoumikhin opened Nov 5, 2026Contributor·

  ·
  12

• Limit OpenMP parallelism in SDPA reference tests

  CLA Signedmeta-exported
  #24651·psiddh opened Nov 5, 2026Contributor·

  ·
  4

• Harden Android instrumentation CI and isolate performance checks (#24650)

  CLA Signedmeta-exported
  #24650·psiddh opened Nov 5, 2026Contributor·

  ·
  6

• Docs: more C++ troubleshooting and a GPU input and output example for the pip package

  CLA Signed
  #24649·pytorchbot opened Nov 5, 2026Collaborator·

  ·
  1

• OpenVINO: document how copies are handled, and test constant copies off the CPU

  CLA Signedrelease notes: openvino
  #24647·shoumikhin opened Nov 5, 2026Contributor·

  ·
  13

• MLX: make the harness signal tests independent of the caller, and give the CPU index ranges for 8-byte dtypes

  CLA Signedrelease notes: mlx
  #24646·shoumikhin opened Nov 5, 2026Contributor·

  ·
  9

• Add the fused_quant op dialect in exir/dialects/fused_quant

  CLA Signedmeta-exported
  #24645·mcremon-meta opened Nov 5, 2026Contributor·

  ·
  3

• Core ML tests: skip the ReadOnCpu case of the uint8 buffer test in fbcode

  CLA Signedmeta-exported
  #24644·digantdesai opened Nov 5, 2026Contributor·

  ·
  4

• Fix CUDA AOTI shim linkage in Buck

  CLA Signedmeta-exported
  #24643·mattiasbrand opened Nov 5, 2026·

  ·
  7

• [Native] Add chunked prefill and interactive chat to the LLM runner

  CLA Signed
  #24639·metascroy opened Nov 5, 2026Contributor·

  ·
  1

• [Native] Run off-graph KV cache packages in the LLM runner

  CLA Signed
  #24638·metascroy opened Nov 5, 2026Contributor·

  ·
  1

• [Native] Serve kvcache::update_and_attend in the ggml engine

  CLA Signed
  #24637·metascroy opened Nov 5, 2026Contributor·

  ·
  1

• [Native] Add load-time options to engine contexts

  CLA Signed
  #24636·metascroy opened Nov 5, 2026Contributor·

  ·
  1

• [Native] Export HF LLMs with an off-graph KV cache

  CLA Signed
  #24635·metascroy opened Nov 5, 2026Contributor·

  ·
  1

• Move off-graph HF attention to extension/llm/cache

  CLA Signed
  #24634·metascroy opened Nov 5, 2026Contributor·

  ·
  2

• [CoreAI] Add Core AI backend docs

  CLA Signed
  #24630·metascroy opened Nov 4, 2026Contributor·

  ·
  2

• Bump minimatch from 3.1.2 to 3.1.5 in /examples/demo-apps/react-native/rnllama

  CLA Signeddependenciesjavascript
  #24629·dependabot[bot] opened Nov 4, 2026Bot·

  ·
  2

• exir: lower rotary embeddings written with complex numbers for every backend

  CLA Signedrelease notes: exir
  #24627·shoumikhin opened Nov 4, 2026Contributor·

  ·
  44

• Transforms: Keep integer scalars exact when replacing them with tensors

  CLA Signedcommunity: contribution
  #24625·abm-77 opened Nov 4, 2026Contributor·

  ·
  3

• Add opt-in prompt-boundary snapshots for LLM serving

  CLA Signed
  #24624·mergennachin opened Nov 4, 2026Contributor·

  ·
  3

• Prefer longer cached prefixes when replacing LLM serving sessions

  CLA Signed
  #24623·mergennachin opened Nov 4, 2026Contributor·

  ·
  6