Class QuantPolicy

java.lang.Object

smile.llm.quant.QuantPolicy

public final class QuantPolicy
extends Object

Resolves checkpoint QuantFormat and WeightGemmBackend for load.

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
static final record	QuantPolicy.Resolved	Resolved quant policy for a checkpoint load.

Method Summary

Modifier and Type	Method	Description
static QuantPolicy.Resolved	resolve(Path checkpointDir, Device device, String backendOverride)	Detects format and selects backend (auto or explicit override).

Methods inherited from class Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Method Details

resolve

public static QuantPolicy.Resolved resolve(Path checkpointDir,
 Device device,
 String backendOverride)
                                    throws IOException

Detects format and selects backend (auto or explicit override).

Parameters:

checkpointDir - model directory.

device - target device.

backendOverride - auto/dense/fp8/nvfp4/marlin, or null to use QuantBackendOverride / auto.

Returns:

resolved format and backend.

Throws:

IOException - if checkpoint IO fails.