mlx-bun library API
    Preparing search index...
    • Port of quantize() in the reference. bits in {2,4,5,8}; signed is only meaningful (and only ever true) for bits===8 (q8_0/int8) — every sub-8-bit key type in QUANT_PARAMS is unsigned. Groups run along the LAST axis (head_dim), BLOCK_SIZE=32 elements per group.

      Parameters

      • x: MlxArray
      • bits: number
      • signed: boolean
      • s: bigint = gpuStream

      Returns TurboQuantKeyEncoded