All pull requests

Pull requests

• [feat]: CompactH3 playground clip-queue (fork only)

  #43·aryan5v opened Sep 19, 2026Owner·

  ·
  2

• [feat]: per-role LoRA for H3 QAD + two-stage QAT recipe

  #42·aryan5v opened Sep 19, 2026Owner·

  ·
  5

• [feat]: CompactH3 recovery, DMD2, and quantization

  #41·aryan5v opened Sep 17, 2026Owner·

  ·
  9

• [perf]: head-shard MiniMax-H3 hybrid attention under sequence parallel

  #40·aryan5v opened Sep 15, 2026Owner·

  ·
  8

• [feat]: Wire creation studio config to backend session init and generation

  #39·aryan5v opened Sep 8, 2026Owner·

  ·
  2

• [feat]: Dreamverse creation studio UI uplift (Dreamina-inspired)

  #38·aryan5v opened Sep 8, 2026Owner·

  ·
  2

• [feat]: MiniMax H3 hybrid window+linear attention

  #35·aryan5v opened Sep 3, 2026Owner·

  ·
  45

• [perf]: dispatch wide-M affine H3 MLX linears through dequant plus dense GEMM

  #34·aryan5v opened Aug 30, 2026Owner·

  ·
  11

• [perf] Add MiniMax H3 MLX VSA and SIMD attention

  #33·aryan5v opened Aug 28, 2026Owner·

  ·
  18

• [feat] Add MiniMax H3 MLX T2VA inference

  #32·aryan5v opened Aug 27, 2026Owner·

  ·
  16

• [new-model] Add LTX-2.5 inference support

  #30·aryan5v opened Aug 12, 2026Owner·

  ·
  12

• [feat]: MLX probes, causal/streaming attention, W8A8, MXFP4 QAT, I2V

  #29·aryan5v opened Aug 11, 2026Owner·

  ·
  1

• 📝 Add docstrings to mlx-runtime-aggregate

  #28·coderabbitai[bot] opened Aug 8, 2026Bot·

  ·
  2

  

• Aggregate MLX runtime: refine / fast-spatial / prompt-enhance + Wan2.2, causal, probes

  #27·aryan5v opened Aug 5, 2026Owner·

  ·
  36

• Release generic MotionKernel artifact dispatch V1

  #26·aryan5v opened Aug 2, 2026Owner·

  ·
  12

• [feat]: capture FX graph metadata during optimization profiles

  #19·aryan5v opened Jul 31, 2026Owner·

  ·
  4

• [feat]: add workload-driven generation launcher

  #18·aryan5v opened Jul 31, 2026Owner·

  ·
  14

• [feat]: complete Wan MotionKernel optimization integration

  #16·aryan5v opened Jul 30, 2026Owner·

  ·
  1

• Add PromptRL training support

  #15·aryan5v opened Jul 27, 2026Owner·

  ·
  26

• [feat]: native Apple MLX studio with Liquid Glass onboarding

  #14·aryan5v opened Jul 10, 2026Owner·

  ·
  11

• [feat]: preserve Apple MLX future 5B, I2V, and self-forcing tracks

  #13·aryan5v opened Jul 9, 2026Owner·

  ·
  6

• [feat]: add MLX INT8 FastWan-QAD 1.3B inference on Apple Silicon

  #12·aryan5v opened Jul 9, 2026Owner·

  ·
  6