WOLFRAM

Wolfram Language & System Documentation Center

NetEvaluator

is an option for certain functions that specifies which software back end should perform the computation.

Details
Details and Options Details and Options
Examples  
Basic Examples  
Properties & Relations  
Possible Issues  
See Also
Tech Notes
Related Workflows
History
Cite this Page

NetEvaluator

NetEvaluator

is an option for certain functions that specifies which software back end should perform the computation.

Details

  • NetEvaluator is typically used when neural networks and layers are evaluated on an input.
  • Possible settings include:
  • Automaticinfer using the TargetDevice value
    "MXNet"use the "MXNet" back end
    "ONNXRuntime"use the "ONNXRuntime" back end
  • The value Automatic is equivalent to "ONNXRuntime" when TargetDevice is resolved to "CoreML" or "DirectML". In all other cases, it is equivalent to "MXNet".

Examples

open all close all

Basic Examples  (1)

Evaluate a network using the "ONNXRuntime" back end:

Wolfram Language code: net = NetInitialize@NetChain[{10, Ramp, 3}, "Input" -> {2}]
Wolfram Language code: net[{1, 2}, NetEvaluator -> "ONNXRuntime"]

Properties & Relations  (1)

Depending on the net and the platform, different back ends can have performance differences:

Wolfram Language code: net = NetModel["CLIP Multi-domain Feature Extractor"]
Wolfram Language code: input = [image];
Wolfram Language code: outMX = net[input, NetEvaluator -> "MXNet"];//RepeatedTiming
Wolfram Language code: outORT = net[input, NetEvaluator -> "ONNXRuntime"];//RepeatedTiming

Results from different back ends can differ by negligible amounts, often due to the numerical precision provided to WorkingPrecision or other rounding errors coming from parallelization:

Wolfram Language code: Max@Abs[outMX - outORT]

Possible Issues  (1)

Evaluation with the "ONNXRuntime" back end fails for nets that cannot be exported to ONNX:

Wolfram Language code: layer = CatenateLayer[1, InputPorts -> {{"Varying", 3}, {2}}]
Wolfram Language code: Export["layer.onnx", layer]
Wolfram Language code: layer[<|"Input1" -> {{1, 2, 3}}, "Input2" -> {4, 5}|>, NetEvaluator -> "ONNXRuntime"]
Wolfram Research (2026), NetEvaluator, Wolfram Language function, https://reference.wolfram.com/language/ref/NetEvaluator.html.

Text

Wolfram Research (2026), NetEvaluator, Wolfram Language function, https://reference.wolfram.com/language/ref/NetEvaluator.html.

CMS

Wolfram Language. 2026. "NetEvaluator." Wolfram Language & System Documentation Center. Wolfram Research. https://reference.wolfram.com/language/ref/NetEvaluator.html.

APA

Wolfram Language. (2026). NetEvaluator. Wolfram Language & System Documentation Center. Retrieved from https://reference.wolfram.com/language/ref/NetEvaluator.html

BibTeX

@misc{reference.wolfram_2026_netevaluator, author="Wolfram Research", title="{NetEvaluator}", year="2026", howpublished="\url{https://reference.wolfram.com/language/ref/NetEvaluator.html}", note=[Accessed: 12-June-2026]}

BibLaTeX

@online{reference.wolfram_2026_netevaluator, organization={Wolfram Research}, title={NetEvaluator}, year={2026}, url={https://reference.wolfram.com/language/ref/NetEvaluator.html}, note=[Accessed: 12-June-2026]}