All pull requests

Pull requests

• Q40 expert projections lacked a routed Metal implementation; add per-…

  #3342·kali opened Nov 10, 2026Collaborator·

  ·
  2

• core: sum a constant quantized matmul operand at codegen, not at run time

  #3337·oysteinkrog opened Nov 9, 2026·

  ·

• pixel lab: RVM, synced output, mask motion, and pipeline stats

  #3335·czoli1976 opened Nov 7, 2026Contributor·

  ·
  2

• examples: Kyutai's Moshi, full duplex over WAVs

  #3333·kali opened Nov 7, 2026Collaborator·

  ·
  3

• wgpu: register-blocked depthwise convolution for vec4-aligned NCHW

  #3318·czoli1976 opened Nov 5, 2026Contributor·

  ·

• wgpu: tile the direct convolution over two output rows in f16

  #3317·czoli1976 opened Nov 5, 2026Contributor·

  ·

• Add opt-in FP16 precision policies to WGPU

  #3316·czoli1976 opened Nov 5, 2026Contributor·

  ·
  1

• examples: Bonsai Image, native and in the browser

  #3306·czoli1976 opened Nov 4, 2026Contributor·

  ·
  6

• Fuse routed Q40 projections with SwiGLU on Metal

  #3298·JulienBalianSonos opened Nov 3, 2026Collaborator·

  ·
  2

• Add expert bias directly to routed Metal outputs

  #3297·JulienBalianSonos opened Nov 3, 2026Collaborator·

  ·
  1

• wgpu: route small direct convolution grids through split-K

  #3285·czoli1976 opened Nov 3, 2026Contributor·

  ·
  3

• cuda: Q3_K, IQ4_XS, and IQ2_S matmul fallback by expanding constants to floats

  #3063·czoli1976 opened Oct 18, 2026Contributor·

  ·

• cuda: Q4_K matmul fallback through a per-run device dequant

  #3062·czoli1976 opened Oct 18, 2026Contributor·

  ·

• cuda: Q5_K matmul fallback through a per-run device dequant

  #3061·czoli1976 opened Oct 18, 2026Contributor·

  ·

• cuda: Q6_K matmul fallback through a per-run device dequant

  #3060·czoli1976 opened Oct 18, 2026Contributor·

  ·

• cuda: Q8_0 matmul fallback through a per-run device dequant

  #3059·czoli1976 opened Oct 18, 2026Contributor·

  ·

• metal: Q3_K, IQ4_XS, and IQ2_S matmul fallback by expanding constants to floats

  #3058·czoli1976 opened Oct 17, 2026Contributor·

  ·

• metal: Q4_K matmul fallback through a per-run device dequant

  #3057·czoli1976 opened Oct 17, 2026Contributor·

  ·

• metal: Q5_K matmul fallback through a per-run device dequant

  #3056·czoli1976 opened Oct 17, 2026Contributor·

  ·

• metal: Q6_K matmul fallback through a per-run device dequant

  #3055·czoli1976 opened Oct 17, 2026Contributor·

  ·

• metal: Q8_0 matmul fallback through a per-run device dequant

  #3054·czoli1976 opened Oct 17, 2026Contributor·

  ·

• core, nnef: install APEX GGUF weights on float matmul constants

  #3053·czoli1976 opened Oct 17, 2026Contributor·

  ·

• nnef: read APEX GGUF tensors into NNEF blobs

  #3052·czoli1976 opened Oct 17, 2026Contributor·

  ·

• core: pack constant APEX weights for arch matmul tiles

  #3051·czoli1976 opened Oct 17, 2026Contributor·

  ·

• linalg: APEX panels for the wasm32 matmul tiles

  #3050·czoli1976 opened Oct 17, 2026Contributor·

  ·