Class Nvfp4Linear

java.lang.Object

smile.llm.quant.Nvfp4Linear

All Implemented Interfaces:

AutoCloseable, LinearOp

public final class Nvfp4Linear
extends Object
implements LinearOp, AutoCloseable

NVFP4 weight-only linear for Blackwell (sm_100+).

Wraps LibTorch / cuBLASLt NVFP4 GEMM when available. Until the linked LibTorch exposes a stable NVFP4 scaled-mm entry point, forward(Tensor) throws a clear error (no Marlin fallback on Blackwell).

Constructor Summary

Constructors

Constructor	Description
Nvfp4Linear(Tensor weight, Tensor weightScale, Tensor bias)	Creates an NVFP4 weight-only linear layer.

Method Summary

Modifier and Type	Method	Description
void	close()
Tensor	forward(Tensor input)	Applies the linear transformation.
int	inFeatures()	Returns the input feature count.
int	outFeatures()	Returns the output feature count.

Methods inherited from class Object

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

Constructor Details

Nvfp4Linear

public Nvfp4Linear(Tensor weight,
 Tensor weightScale,
 Tensor bias)

Creates an NVFP4 weight-only linear layer.

Parameters:

weight - NVFP4 weight tensor.

weightScale - weight scale tensor.

bias - optional bias, or null.

Method Details

inFeatures

public int inFeatures()

Returns the input feature count.

Returns:

input features.

outFeatures

public int outFeatures()

Returns the output feature count.

Returns:

output features.

forward

public Tensor forward(Tensor input)

Description copied from interface: LinearOp

Applies the linear transformation.

Specified by:

forward in interface LinearOp

Parameters:

input - activations (caller retains ownership).

Returns:

owned output tensor (caller must close).

close

public void close()

Specified by:

close in interface AutoCloseable