Issues

• perf: Optimize Qwen3.8-27B DFlash2 execution using lessons from Splash

  mlx-node/mlx-node#161 · Filter by author Brooooooklyn opened on Sep 19, 2026

• Performance tracker: mlx-node vs llama.cpp across models (prefill & decode)

  mlx-node/mlx-node#142 · Filter by author Brooooooklyn opened on Sep 9, 2026
  2

• Optimize for Metal 4.1

  mlx-node/mlx-node#135 · Filter by author Brooooooklyn opened on Sep 1, 2026

• Native FP8/FP4 execution on DGX Spark: qqmm is unbound and fp8_e4m3 has no kernel

  enhancement
  mlx-node/mlx-node#134 · Filter by author Brooooooklyn opened on Sep 1, 2026
  2

• model_type "qwen3_moe" is rejected by loadModel()

  mlx-node/mlx-node#91 · Filter by author FrankDaze opened on Jul 8, 2026