Function analyzeSensitivityExact

analyzeSensitivityExact(
    model: RuntimeModel,
    layers: Map<string, QuantizedLinear>,
    calIds: number[][],
    options?: AnalyzeSensitivityOptions,
    onProgress?: (done: number, total: number, layerName: string) => void,
): SensitivityResult[]
Per-layer KL sensitivity of an already-quantized MLX model.
Parameters
- model: RuntimeModel
  loaded RuntimeModel (its QuantizedLinears are the uniform baseline; their dequantized weights are the self-contained bf16 source).
- layers: Map<string, QuantizedLinear>
  the quantizable linears to probe, keyed by module path (typically model.loraTargets()).
- calIds: number[][]
  calibration samples — each a row of int32 token ids.
- options: AnalyzeSensitivityOptions = {}
  candidateBits (default [4,8]) and groupSize (default 64).
- OptionalonProgress: (done: number, total: number, layerName: string) => void
Returns SensitivityResult[]
- Defined in packages/quantize/src/sensitivity.ts:141