mlx-bun library API
    Preparing search index...

    Fixed M=1..gamma signed-Q8 kernel family for native MTP draft and accepted token absorption. The row dot-product and materialization boundaries are identical for every batch width; only grid.z changes.

    Hierarchy

    • Glm52CanonicalMetalExecutor
      • Glm52CanonicalQ8MetalExecutor
    Index