All pull requests

Pull requests

• finegrained-kernels: add FP8 weight-only inference

  chore
  #1241·IlyasMoutawwakil opened Oct 16, 2026Member·3 of 4 tasks·

  ·
  1

• finegrained-kernels: add rowwise FP8 across GEMMs and MoE

  chore
  #1240·IlyasMoutawwakil opened Oct 16, 2026Member·3 of 4 tasks·

  ·
  1

• finegrained-kernels: add quantized embedding lookup

  chore
  #1239·IlyasMoutawwakil opened Oct 16, 2026Member·3 of 4 tasks·

  ·
  1

• finegrained-kernels: accept MX scale bytes without native UE8M0

  chore
  #1238·IlyasMoutawwakil opened Oct 16, 2026Member·3 of 4 tasks·

  ·
  2

• finegrained-kernels: fix tuner selection and expose phase controls

  chore
  #1237·IlyasMoutawwakil opened Oct 16, 2026Member·2 of 3 tasks·

  ·
  2

• finegrained-kernels: decode losslessly compressed NVFP4 scales (CSF)

  chore
  #1236·IlyasMoutawwakil opened Oct 16, 2026Member·3 of 4 tasks·

  ·
  1

• feat: stable abi and support torch compile

  feature
  #1234·drbh opened Oct 15, 2026Collaborator·

  ·
  10

• deep-ep: add DeepEP 2.5.0 (expert-parallel dispatch/combine)

  chore
  #1231·IlyasMoutawwakil opened Oct 15, 2026Member·3 of 6 tasks·

  ·
  12

• fix(ci): harden workflow triggers, secrets and gates (#1227)

  fix
  #1228·hf-security-analysis[bot] opened Oct 14, 2026Bot·

  ·

• build(deps): bump the actions group across 1 directory with 7 updates

  dependenciesdepsgithub_actions
  #1227·dependabot[bot] opened Oct 14, 2026Bot·

  ·

• Fa2 stable abi api align

  #1226·drbh opened Oct 14, 2026Collaborator·

  ·
  7

• fla: make compile friendly

  chore
  #1223·vasqu opened Oct 13, 2026Collaborator·

  ·
  35

• ragged-dot-tpu: add tokamax MoE experts kernel for TPU

  chore
  #1208·tengomucho opened Oct 8, 2026Member·3 of 6 tasks·

  ·
  7

• megablocks: add smoke tests and convert to stable ABI

  chore
  #1199·danieldk opened Oct 2, 2026Member·

  ·
  5

• deformable-detr: convert to Torch Stable ABI and add tests

  chore
  #1198·danieldk opened Oct 1, 2026Member·

  ·
  9

• gemma-quant-kernels: add fused INT2/INT4/INT8 GEMV, GEMM, and MoE grouped GEMM

  chore
  #1192·pculliton opened Sep 29, 2026·3 of 6 tasks·

  ·
  9

• cv-utils: add resize_normalize and resize_normalize_patchify Triton kernels

  chore
  #1182·molbap opened Sep 25, 2026·

  ·
  11

• fix(ci): harden GitHub Actions workflows (#1169)

  fix
  #1170·hf-security-analysis[bot] opened Sep 22, 2026Bot·

  ·

• finegrained-fp8: bound the grouped kernels to N and K

  chore
  #1154·malaiwah opened Sep 16, 2026·

  ·
  1

  5

• rotary: support partial RoPE in apply_rotary_transformers

  chore
  #1143·piyushumate opened Sep 9, 2026·

  ·
  24

• fla: sync with flash-linear-attention main

  chore
  #1141·kashif opened Sep 7, 2026Contributor·2 of 6 tasks·

  ·
  1

• chore: update pr-comment-build.yaml

  chore
  #1114·hf-security-analysis[bot] opened Aug 27, 2026Bot·

  ·

• sgl-flash-attn3: switch to torch stable ABI

  chore
  #1054·adarshxs opened Jul 31, 2026Contributor·6 tasks·

  ·
  23

• vllm-quantization: import in kernels-community again

  cudafeaturenew-kernelrocm
  #991·danieldk opened Jun 23, 2026Member·

  ·

• tokenspeed-attention: add TokenSpeed MHA/MLA Triton kernels

  featurenew-kerneltritonupstream-sync
  #990·adarshxs opened Jun 23, 2026Contributor·

  ·
  2