public class QuantizationOptions
Options for deterministic, row-local symmetric Q8 conversion. Source matrices become physical INT8; other tensors retain their original encoding. This is weight-only quantization, not activation calibration.
options = QuantizationOptions(groupSize=128l, verbose=true)
| Modifier and Type | Member | Description |
|---|---|---|
| public readonly property | bufferBytesbufferBytes as Long |
Source block budget in bytes, independent of total model size. |
| public readonly property | groupSizegroupSize as Long |
Scalars per row-local group; power of two in 16..4096. |
| public readonly property | verboseverbose as Boolean |
Whether to display bounded-rate console progress. |
| Modifier and Type | Member | Description |
|---|---|---|
| public | QuantizationOptionsQuantizationOptions(groupSize as Long = 128l, bufferBytes as Long = 1048576l, verbose as Boolean = true) |
Configures Q8 grouping and the bounded-memory conversion pipeline. |