All pull requests

Pull requests

• feat(msa): add pooled-key decode block scores

  #30·yechank-nvidia opened Nov 5, 2026·

  ·

• feat(msa): add pooled-key prefill block scores

  #29·yechank-nvidia opened Nov 5, 2026·

  ·

• q8kv4 decode: opt-in PDL for the decode indexer and decode attention

  #28·Orien-Xu opened Oct 24, 2026·

  ·

• fmha_sm100(cute): do not depend on quack.activation.sub_packed_f32x2

  #27·Orien-Xu opened Oct 24, 2026·

  ·
  1

• jit: recover from torch extension locks left by killed builders

  #26·Orien-Xu opened Oct 24, 2026·

  ·

• Q8KV4 prefill: build for SM107, speed up K1, unify the combine, place partials by locality domain

  #21·pgera opened Oct 22, 2026·

  ·

• feat: support NHD paged KV cache layout via kv_layout="NHD"

  #14·alumkal opened Aug 13, 2026·

  ·

• fix: support external page tables with 4-byte alignment

  #12·rogeryoungh opened Aug 6, 2026Collaborator·

  ·

• Add SM12x MSA Support (Tested on DGX Spark, linux/arm64/SM121)

  #1·dbotwinick opened Jun 11, 2026·

  ·