WOLFRAM

Wolfram Language & System Documentation Center
Wolfram Language Home Page »

NetExternalObject
✖
NetExternalObject

Updatedshow changeshide changes[Experimental]

✖

represents a net model in an external framework format.

Details

  • NetExternalObject is typically used to run inference using neural network models not defined in the Wolfram Language.
  • Supported file formats are:
  • "ONNX"ONNX models
    "MXNet"MXNet models
    "GGUF"GGUF models
  • For a model with a single input port, NetExternalObject[…][data] gives the result of applying the model to data.
  • For a model with multiple input ports, NetExternalObject[…][<|port1data1,…|>] provides datai to porti.
  • When data is given as NumericArray, the output will be a NumericArray. »
  • Information[NetExternalObject[…]] generates an information panel about the model. »
  • Information of a NetExternalObject may include the following properties:
  • "Backend"the evaluation back end used by the model
    "ContextWindowSize"current maximum size of the model context windows
    "InputPortNames"list of names of input ports
    "InputPorts"association of input port shapes
    "OutputPortNames"list of names of output ports
    "OutputPorts"association of output port shapes
    "Properties"available properties
    "RawData"unprocessed native contents of the model
    "TrainingContextWindowSize"original size of the model context windows
  • NetExternalObject[…][data,opts] specifies that options should be used in applying the net to data. Possible options include:
  • Method Automaticspecify details of the model evaluation
    TargetDevice "CPU"the device on which to perform evaluation
  • Possible settings for TargetDevice change depending on the system and the hardware and include:
  • "CPU"use the CPU
    "GPU"use available dedicated hardware
    "CoreML"Apple CoreML framework
    "CUDA"Nvidia CUDA API
    "DirectML"Microsoft Direct Machine Learning API
  • For a NetExternalObject with LLaMA.cpp backend ("GGUF" format) that generates text, option Method can be set to a list of rules having any of the following keys:
  • "MaxTokens"Automaticmaximum number of generated tokens
    RandomSeedingAutomaticwhat seeding of pseudorandom generators should be done internally
    "PrependPrompt"Falsewhether to prepend the input to ext to the output text
    "ResetState"Truewhether to reset the language model's state before processing the input
    "StopTokens"{}a list of strings that should stop the generation if produced

Examples

open allclose all

Basic Examples  (2)Summary of the most common use cases

Import an ONNX model as an external network:

In[1]:=1
✖
https://wolfram.com/xid/0dc2m79gcrde-kvn6hn
Out[1]=1

Evaluate the model on some input data:

In[2]:=2
✖
https://wolfram.com/xid/0dc2m79gcrde-gjt1mz
Out[2]=2

Import a quantized image classifier:

In[1]:=1
✖
https://wolfram.com/xid/0dc2m79gcrde-digatt
Out[1]=1

Define a minimal pipeline to pre- and post-process the data:

In[2]:=2
✖
https://wolfram.com/xid/0dc2m79gcrde-pk8cc2

Run the external model:

In[3]:=3
✖
https://wolfram.com/xid/0dc2m79gcrde-4m1w6f
Out[3]=3

Scope  (5)Survey of the scope of standard use cases

Formats  (3)

Import an ONNX model as a NetExternalObject:

In[1]:=1
✖
https://wolfram.com/xid/0dc2m79gcrde-j8xvci
Out[1]=1

Import a net in GGUF format (model credit: Olusegun Odewole, https://huggingface.co/segestic/Tinystories-gpt-0.1-3m ):

In[1]:=1
✖
https://wolfram.com/xid/0dc2m79gcrde-two0nh
Out[1]=1

Import an MXNet model as a NetExternalObject:

In[1]:=1
✖
https://wolfram.com/xid/0dc2m79gcrde-9m6p82
Out[1]=1

Properties  (2)

Extract a property of the external net using Information:

In[1]:=1
✖
https://wolfram.com/xid/0dc2m79gcrde-1mckga
Out[1]=1
In[2]:=2
✖
https://wolfram.com/xid/0dc2m79gcrde-7fwug3
Out[2]=2

Get a list of properties:

In[3]:=3
✖
https://wolfram.com/xid/0dc2m79gcrde-miq8vx
Out[3]=3

Get a list of all the supported properties:

In[1]:=1
✖
https://wolfram.com/xid/0dc2m79gcrde-w7gxq5
Out[1]=1
In[2]:=2
✖
https://wolfram.com/xid/0dc2m79gcrde-m49rw1
Out[2]=2

Options  (4)Common values & functionality for each option

Method  (2)

Import a text-generating model:

In[1]:=1
✖
https://wolfram.com/xid/0dc2m79gcrde-5xp07o
Out[1]=1

Run the model until a termination token is reached:

In[9]:=9
✖
https://wolfram.com/xid/0dc2m79gcrde-psbl0c
Out[9]=9

Prepend the prompt to the generated text:

In[10]:=10
✖
https://wolfram.com/xid/0dc2m79gcrde-um2s1k
Out[10]=10

Specify a maximum amount of tokens to generate:

In[28]:=28
✖
https://wolfram.com/xid/0dc2m79gcrde-yo3i3j
Out[28]=28

Text generation is a random process:

In[1]:=1
✖
https://wolfram.com/xid/0dc2m79gcrde-vg21uv
In[2]:=2
✖
https://wolfram.com/xid/0dc2m79gcrde-r51k20
Out[2]=2

Fix the seed of the pseudorandom generator to get reproducible results:

In[3]:=3
✖
https://wolfram.com/xid/0dc2m79gcrde-cfnpu8
Out[3]=3

TargetDevice  (2)

Import an ONNX file and run it on the CPU:

In[1]:=1
✖
https://wolfram.com/xid/0dc2m79gcrde-mdr267
Out[1]=1
In[2]:=2
✖
https://wolfram.com/xid/0dc2m79gcrde-1i1fzy
Out[2]=2

Run the model using CoreML:

In[3]:=3
✖
https://wolfram.com/xid/0dc2m79gcrde-cmg72y
Out[3]=3

Some models will run faster when evaluated on specialized hardware:

In[1]:=1
✖
https://wolfram.com/xid/0dc2m79gcrde-nhn8l3
Out[1]=1

Timing on CPU:

In[2]:=2
✖
https://wolfram.com/xid/0dc2m79gcrde-fsvea
Out[2]=2

Timing on CoreML:

In[3]:=3
✖
https://wolfram.com/xid/0dc2m79gcrde-vf8x5t
Out[3]=3

Applications  (3)Sample problems that can be solved with this function

Run a real-time face detector:

In[1]:=1
✖
https://wolfram.com/xid/0dc2m79gcrde-x6a9rn
In[2]:=2
✖
https://wolfram.com/xid/0dc2m79gcrde-xn5259
Out[2]=2

Define a transformation from an image to a list of arrays:

In[3]:=3
✖
https://wolfram.com/xid/0dc2m79gcrde-ppneid
Out[3]=3

Run the detector on the encoded image:

In[4]:=4
✖
https://wolfram.com/xid/0dc2m79gcrde-ybqoc0
Out[4]=4

Extract the bounding box and probability information from the result:

In[5]:=5
✖
https://wolfram.com/xid/0dc2m79gcrde-nzrkfp

Visualize the detections, filtering the ones with low probability or large overlap:

In[7]:=7
✖
https://wolfram.com/xid/0dc2m79gcrde-m0p94x
Out[7]=7

This network can analyze more that 40 images per second on a consumer laptop:

In[9]:=9
✖
https://wolfram.com/xid/0dc2m79gcrde-ze06zs
Out[9]=9

Import a text-generating model:

In[1]:=1
✖
https://wolfram.com/xid/0dc2m79gcrde-h2z6hb
Out[1]=1

Run the model until a termination token is reached:

In[2]:=2
✖
https://wolfram.com/xid/0dc2m79gcrde-36vv3m
Out[2]=2

Import a "GGUF" file as an embedding model:

In[1]:=1
✖
https://wolfram.com/xid/0dc2m79gcrde-0t2tv1
Out[1]=1

Generate an embedding vector for each token in a string:

In[2]:=2
✖
https://wolfram.com/xid/0dc2m79gcrde-89jtxs

Properties & Relations  (2)Properties of the function, and connections to other functions

Input given as a NumericArray will cause the output to also be a NumericArray:

In[1]:=1
✖
https://wolfram.com/xid/0dc2m79gcrde-6tn454
Out[1]=1
In[2]:=2
✖
https://wolfram.com/xid/0dc2m79gcrde-nrsli9
Out[2]=2

External net models typically have a hard-coded batch dimension:

In[1]:=1
✖
https://wolfram.com/xid/0dc2m79gcrde-g2ycdk
Out[1]=1
In[2]:=2
✖
https://wolfram.com/xid/0dc2m79gcrde-5jjp6f
Out[2]=2

Ordinary networks do not have an explicit batch dimension:

In[3]:=3
✖
https://wolfram.com/xid/0dc2m79gcrde-yxwp34
Out[3]=3

They work both on single examples and batches of any size:

In[4]:=4
✖
https://wolfram.com/xid/0dc2m79gcrde-p1gz81
Out[4]=4
In[5]:=5
✖
https://wolfram.com/xid/0dc2m79gcrde-7w3n3s
Out[5]=5
In[6]:=6
✖
https://wolfram.com/xid/0dc2m79gcrde-uchwzw
Out[6]=6

Possible Issues  (2)Common pitfalls and unexpected behavior

External models cannot be modified:

In[1]:=1
✖
https://wolfram.com/xid/0dc2m79gcrde-3oft4q
Out[1]=1
In[2]:=2
✖
https://wolfram.com/xid/0dc2m79gcrde-nl5yp6
Out[2]=2

External models cannot be retrained:

In[1]:=1
✖
https://wolfram.com/xid/0dc2m79gcrde-5aq730
Out[1]=1
In[2]:=2
✖
https://wolfram.com/xid/0dc2m79gcrde-klc21x
Out[2]=2
Wolfram Research (2022), NetExternalObject, Wolfram Language function, https://reference.wolfram.com/language/ref/NetExternalObject.html (updated 2025).
✖
Wolfram Research (2022), NetExternalObject, Wolfram Language function, https://reference.wolfram.com/language/ref/NetExternalObject.html (updated 2025).

Text

Wolfram Research (2022), NetExternalObject, Wolfram Language function, https://reference.wolfram.com/language/ref/NetExternalObject.html (updated 2025).

✖
Wolfram Research (2022), NetExternalObject, Wolfram Language function, https://reference.wolfram.com/language/ref/NetExternalObject.html (updated 2025).

CMS

Wolfram Language. 2022. "NetExternalObject." Wolfram Language & System Documentation Center. Wolfram Research. Last Modified 2025. https://reference.wolfram.com/language/ref/NetExternalObject.html.

✖
Wolfram Language. 2022. "NetExternalObject." Wolfram Language & System Documentation Center. Wolfram Research. Last Modified 2025. https://reference.wolfram.com/language/ref/NetExternalObject.html.

APA

Wolfram Language. (2022). NetExternalObject. Wolfram Language & System Documentation Center. Retrieved from https://reference.wolfram.com/language/ref/NetExternalObject.html

✖
Wolfram Language. (2022). NetExternalObject. Wolfram Language & System Documentation Center. Retrieved from https://reference.wolfram.com/language/ref/NetExternalObject.html

BibTeX

@misc{reference.wolfram_2025_netexternalobject, author="Wolfram Research", title="{NetExternalObject}", year="2025", howpublished="\url{https://reference.wolfram.com/language/ref/NetExternalObject.html}", note=[Accessed: 07-June-2025 ]}

✖
@misc{reference.wolfram_2025_netexternalobject, author="Wolfram Research", title="{NetExternalObject}", year="2025", howpublished="\url{https://reference.wolfram.com/language/ref/NetExternalObject.html}", note=[Accessed: 07-June-2025 ]}

BibLaTeX

@online{reference.wolfram_2025_netexternalobject, organization={Wolfram Research}, title={NetExternalObject}, year={2025}, url={https://reference.wolfram.com/language/ref/NetExternalObject.html}, note=[Accessed: 07-June-2025 ]}

✖
@online{reference.wolfram_2025_netexternalobject, organization={Wolfram Research}, title={NetExternalObject}, year={2025}, url={https://reference.wolfram.com/language/ref/NetExternalObject.html}, note=[Accessed: 07-June-2025 ]}