Function convertToAffine

convertToAffine<C extends Cache>(
    source: {
        offset: number;
        toQuantized(groupSize: number, bits: number, kernels: AffineKernelSet): C;
    },
    groupSize: number,
    bits: number,
    start: number,
    kernelSet: AffineKernelSet,
): C
The affine conversion of one bf16 entry: the source's own toQuantized (which consumes the source) with kernelSet, the set the converted storage attends with, then the shared bookkeeping.
Type Parameters
- C extends Cache
Parameters
- source: {
  offset: number;
  toQuantized(groupSize: number, bits: number, kernels: AffineKernelSet): C;
  }
- groupSize: number
- bits: number
- start: number
- kernelSet: AffineKernelSet
Returns C
- Defined in packages/inference/src/state/bf16-first-kv.ts:55