All pull requests

Pull requests

• [executorch][cuda] Add triton::convolution and the conv routing framework

  CLA Signed
  #24982·Gasoonjia opened Nov 12, 2026Contributor·

  ·
  2

• [executorch][cuda] Add planned replacement to the Triton replacement pass

  CLA Signed
  #24981·Gasoonjia opened Nov 12, 2026Contributor·

  ·
  2

• [executorch][cuda] Deprecate triton_kernel_mode=OFF

  CLA Signed
  #24980·Gasoonjia opened Nov 12, 2026Contributor·

  ·
  2

• Size Muse Glimmer sliding-window ring buffer from the prefill chunk

  CLA Signed
  #24979·Gasoonjia opened Nov 12, 2026Contributor·

  ·
  3

• Arm backend: Delegate int64 ops with narrower integer operands under EXT-INT64

  ciflow/trunkCLA Signedmodule: armpartner: armrelease notes: arm
  #24978·abm-77 opened Nov 12, 2026Contributor·

  ·
  11

• Bound Samsung device allocation waits and clean up queued requests

  #24977·pk954061458 opened Nov 12, 2026·

  ·
  1

  4

• [executorch][vulkan] Fetch embedding_q4gsw indices of rank 2 and 3 from textures

  CLA Signedmeta-exportedmodule: vulkan
  #24976·digantdesai opened Nov 12, 2026Contributor·

  ·
  2

• [executorch][native] Keep float scales of int4 ggml embeddings exact

  CLA Signed
  #24975·digantdesai opened Nov 12, 2026Contributor·

  ·
  2

• [executorch][native] Record embedding op samples

  CLA Signedmeta-exported
  #24974·digantdesai opened Nov 12, 2026Contributor·

  ·
  2

• [executorch][native] Quantize nn.Embedding weights in the op sample schemes

  CLA Signedmeta-exported
  #24973·digantdesai opened Nov 12, 2026Contributor·

  ·
  2

• Keep a bf16 NVFP4 linear's output in bf16 on MLX

  CLA Signed
  #24972·aditvenk opened Nov 12, 2026·

  ·
  4

• Do not delegate quantized linears that have no Vulkan kernel (#24971)

  CLA Signedmeta-exportedmodule: vulkan
  #24971·psiddh opened Nov 12, 2026Contributor·

  ·
  5

• [Native] Launch the Muse Glimmer server on the native backend

  CLA Signed
  #24957·metascroy opened Nov 12, 2026Contributor·

  ·
  1

• [Native] Serve Muse Glimmer on the native ggml engine

  CLA Signed
  #24956·metascroy opened Nov 12, 2026Contributor·

  ·
  1

• [Native] Publish the context length with off-graph packages

  CLA Signed
  #24955·metascroy opened Nov 12, 2026Contributor·

  ·
  1

• [Native] Share the off-graph cache geometry with other callers

  CLA Signed
  #24954·metascroy opened Nov 12, 2026Contributor·

  ·
  1

• Serve one batching session from a single-sequence cache

  CLA Signed
  #24953·metascroy opened Nov 12, 2026Contributor·

  ·
  1

• Report a single-sequence cache's length on SequenceControl

  CLA Signed
  #24952·metascroy opened Nov 12, 2026Contributor·

  ·
  1

• [Native] Export Muse Glimmer through the native backend

  CLA Signed
  #24951·metascroy opened Nov 12, 2026Contributor·

  ·
  1

• [Native] Replay sample sets without decoding their tokens

  CLA Signed
  #24950·metascroy opened Nov 12, 2026Contributor·

  ·
  1

• [Native] Map Muse Glimmer GGUF tensors onto its HF model

  CLA Signed
  #24949·metascroy opened Nov 12, 2026Contributor·

  ·
  1

• [Native] Run GGUF Q5_K weights in the ggml engine

  CLA Signed
  #24948·metascroy opened Nov 12, 2026Contributor·

  ·
  1

• [Native] Load the first layers of a GGUF with --layers

  CLA Signed
  #24947·metascroy opened Nov 12, 2026Contributor·

  ·
  1

• [Native] Run the HF LLM export on newer transformers

  CLA Signed
  #24946·metascroy opened Nov 12, 2026Contributor·

  ·
  1

• [Native] Fuse broadcast RoPE frequencies and lower tanh in ggml

  CLA Signed
  #24945·metascroy opened Nov 12, 2026Contributor·

  ·
  1