public final class MLCExecutionOptions
extends java.lang.Object
A bitmask that specifies the options you’ll use when executing a graph.
| Modifier and Type | Field and Description |
|---|---|
static long |
ForwardForInference
The option to execute the forward pass for inference only.
|
static long |
None |
static long |
PerLayerProfiling
The option to enable additional per layer profiling information currently emitted using signposts.
|
static long |
Profiling
The option to return profiling information in the callback before returning from execution.
|
static long |
SkipWritingInputDataToDevice
The option to skip writing input data to device memory.
|
static long |
Synchronous
The option to execute the graph synchronously.
|
public static final long None
public static final long SkipWritingInputDataToDevice
this option to prevent writing the input tensors to device memory associated with these tensors when the framework executes the graph.
public static final long Synchronous
Include this option to wait until execution of the graph on specified device finishes before returning from the \p execute method.
public static final long Profiling
Include this option to return profliling information in the graph execute completion handler callback, including device execution time.
public static final long ForwardForInference
If you include this option and execute a training graph using one of the \p execute methods, such as \p -executeWithInputsData:lossLabelsData:lossLabelWeightsData:batchSize:options:completionHandler: , the framework only executes the forward pass of the training graph, and it executes that forward pass for inference only.
If you include this option and execute a training graph using one of the executeForward methods, such as \p -executeForwardWithBatchSize:options:completionHandler:), the framework executes the forward pass for inference only.
public static final long PerLayerProfiling
The option to enable per layer profiling information emitted as signposts. The per layer information can be visualized using the Logging Instrument in Xcode's Instruments. This information may not be available for all MLCDevice.