Package com.gencior.triton.core

Class InferRequestedOutput

java.lang.Object
com.gencior.triton.core.InferRequestedOutput
public final class InferRequestedOutput extends Object
Specifies an output tensor to be returned from an inference request.

By default, Triton returns all outputs defined in a model's configuration. Use this class to request only specific outputs, reducing network bandwidth and memory usage — especially useful for models with many output tensors where only a subset is needed.

Usage — Simple (name only):


 InferResult result = client.infer("bert", inputs,
     List.of(InferRequestedOutput.of("embeddings")));
 // Only "embeddings" is returned, other outputs are skipped

Usage — With parameters:


 InferRequestedOutput output = new InferRequestedOutput.Builder("classification")
     .addParameter("classification", 3L) // top-3 classes
     .build();
 InferResult result = client.infer("classifier", inputs, List.of(output));
Since:
1.0.0
Author:
sachachoumiloff

  • Method Details

    • of

      public static InferRequestedOutput of(String name)
      Creates a requested output with just a name and no additional parameters.
      Parameters:
      name - the output tensor name (must match the model's output name)
      Returns:
      a new InferRequestedOutput

    • getName

      public String getName()
      Returns the output tensor name.
      Returns:
      the name

    • getParameters

      public Map<String,Object> getParameters()
      Returns the optional parameters for this output.
      Returns:
      an unmodifiable map of parameters (may be empty)

    • hasParameters

      public boolean hasParameters()
      Returns whether this output has any parameters set.
      Returns:
      true if parameters are present