GitHub Repository: ibm/watson-machine-learning-samples
Path: blob/master/cloud/notebooks/python_sdk/converters/Use ONNX model converted from TensorFlow to recognize hand-written digits.ipynb
⁹³³⁹ views

Kernel: .venv_watsonx_ai_samples_py_312

Use ONNX model converted from TensorFlow to recognize hand-written digits with `ibm-watsonx-ai`

This notebook facilitates ONNX, Tensorflow (and TF.Keras), and watsonx.ai Runtime service. It contains steps and code to work with ibm-watsonx-ai library available in PyPI repository in order to convert the TensorFlow model to ONNX format. It also introduces commands for getting model and training data, persisting model, deploying model and scoring it.

Some familiarity with Python is helpful. This notebook uses Python 3.12.

Learning goals

The learning goals of this notebook are:

Download an externally trained TensorFlow model with dataset.
Convert TensorFlow model to ONNX format
Persist converted model in watsonx.ai Runtime repository.
Deploy model for online scoring using client library.
Score sample records using client library.

This notebook contains the following parts:

1. Setting up the environment

Before you use the sample code in this notebook, you must perform the following setup tasks:

Create a watsonx.ai Runtime instance (information on service plans and further reading can be found here).

Install dependencies

Note: ibm-watsonx-ai documentation can be found here.

In [1]:

%pip install wget | tail -n 1
%pip install matplotlib | tail -n 1
%pip install -U ibm-watsonx-ai | tail -n 1
%pip install "tensorflow==2.18" | tail -n 1
%pip install "tf2onnx==1.16" | tail -n 1
%pip install "onnxruntime==1.21.0" | tail -n 1
%pip install "numpy>=1.26.4,<2" | tail -n 1

Out[1]:

Successfully installed wget-3.2
Successfully installed contourpy-1.3.3 cycler-0.12.1 fonttools-4.61.1 kiwisolver-1.4.9 matplotlib-3.10.8 numpy-2.4.2 pillow-12.1.1 pyparsing-3.3.2
Successfully installed anyio-4.12.1 cachetools-7.0.1 certifi-2026.1.4 charset_normalizer-3.4.4 h11-0.16.0 httpcore-1.0.9 httpx-0.28.1 ibm-cos-sdk-2.14.3 ibm-cos-sdk-core-2.14.3 ibm-cos-sdk-s3transfer-2.14.3 ibm-watsonx-ai-1.5.2 idna-3.11 jmespath-1.0.1 lomond-0.3.3 pandas-2.3.3 pytz-2025.2 requests-2.32.5 tabulate-0.9.0 typing_extensions-4.15.0 tzdata-2025.3 urllib3-2.6.3
Successfully installed absl-py-2.4.0 astunparse-1.6.3 flatbuffers-25.12.19 gast-0.7.0 google-pasta-0.2.0 grpcio-1.78.0 h5py-3.15.1 keras-3.13.2 libclang-18.1.1 markdown-3.10.2 markdown-it-py-4.0.0 markupsafe-3.0.3 mdurl-0.1.2 ml-dtypes-0.4.1 namex-0.1.0 numpy-2.0.2 opt-einsum-3.4.0 optree-0.18.0 protobuf-5.29.6 rich-14.3.2 tensorboard-2.18.0 tensorboard-data-server-0.7.2 tensorflow-2.18.0 termcolor-3.3.0 werkzeug-3.1.5 wheel-0.46.3 wrapt-2.1.1
Successfully installed onnx-1.17.0 protobuf-3.20.3 tf2onnx-1.16.0
Successfully installed coloredlogs-15.0.1 humanfriendly-10.0 mpmath-1.3.0 onnxruntime-1.21.0 sympy-1.14.0
Successfully installed numpy-1.26.4

Connection to watsonx.ai Runtime

Authenticate the watsonx.ai Runtime service on IBM Cloud. You need to provide platform api_key and instance location.

You can use IBM Cloud CLI to retrieve platform API Key and instance location.

API Key can be generated in the following way:

ibmcloud login
ibmcloud iam api-key-create API_KEY_NAME

In result, get the value of api_key from the output.

Location of your watsonx.ai Runtime instance can be retrieved in the following way:

ibmcloud login --apikey API_KEY -a https://cloud.ibm.com
ibmcloud resource service-instance INSTANCE_NAME

In result, get the value of location from the output.

Tip: Your Cloud API key can be generated by going to the Users section of the Cloud console. From that page, click your name, scroll down to the API Keys section, and click Create an IBM Cloud API key. Give your key a name and click Create, then copy the created key and paste it below. You can also get a service specific url by going to the Endpoint URLs section of the watsonx.ai Runtime docs. You can check your instance location in your watsonx.ai Runtime Service instance details.

You can also get service specific apikey by going to the Service IDs section of the Cloud Console. From that page, click Create, then copy the created key and paste it below.

Action: Enter your url and api_key in the following cell.

In [2]:

import getpass

from ibm_watsonx_ai import Credentials

credentials = Credentials(
    url="https://us-south.ml.cloud.ibm.com",
    api_key=getpass.getpass("Please enter your watsonx.ai api key (hit enter): "),
)

In [3]:

from ibm_watsonx_ai import APIClient

client = APIClient(credentials)

1.3. Working with spaces

First of all, you need to create a space that will be used for your work. If you do not have space already created, you can use Deployment Spaces Dashboard to create one.

Click New Deployment Space
Create an empty space
Select Cloud Object Storage
Select watsonx.ai Runtime instance and press Create
Copy space_id and paste it below

Tip: You can also use the ibm_watsonx_ai SDK to prepare the space for your work. More information can be found here.

Action: Assign space ID below

In [4]:

space_id = "ENTER YOUR SPACE ID HERE"

You can use list method to print all existing spaces.

In [ ]:

client.spaces.list(limit=10)

To be able to interact with all resources available in watsonx.ai Runtime, you need to set space which you will be using.

In [5]:

client.set.default_space(space_id)

Out[5]:

'SUCCESS'

2. Downloading externally created TensorFlow model and data

In this section, you will download externally created TensorFlow models and data used for training it. You can choose to download either a TensorFlow or a Keras model.

2.1. Downloading dataset

In [6]:

from pathlib import Path

data_dir = Path("MNIST_DATA")
if not data_dir.is_dir():
    data_dir.mkdir()

In [7]:

import wget

data_path = data_dir / "mnist.npz"
if not data_path.is_file():
    wget.download("https://s3.amazonaws.com/img-datasets/mnist.npz", out=str(data_dir))

In [8]:

import numpy as np

dataset = np.load(data_path)
x_test = dataset["x_test"]

2.2. Downloading TensorFlow model

In [9]:

tf_model_name = "mnist-tf-model"
model_tar_name = f"{tf_model_name}.tar.gz"
model_path = data_dir / model_tar_name

if not model_path.is_file():
    wget.download(
        f"https://github.com/IBM/watsonx-ai-samples/raw/master/cloud/models/tensorflow/mnist/{model_tar_name}",
        out=str(data_dir),
    )

3. Converting TensorFlow model to ONNX format

In this section, you will unpack externally created TensorFlow models, provided in either SavedModel or Keras format (depending on your earlier selection), from the tar archive and convert them to the ONNX format. More information can be found here.

In [10]:

onnx_model_name = "tf_model.onnx"

3.1. Converting TensorFlow SavedModel

In [11]:

import os
import tarfile

os.makedirs(tf_model_name, exist_ok=True)

tar_path = f"MNIST_DATA/{model_tar_name}"
with tarfile.open(tar_path, "r:gz") as tar:
    tar.extractall(path=tf_model_name, filter="data")

Note: If you are working with TensorFlow Lite models, make sure to use the --tflite flag instead of --saved-model.

Tip: You can also run this command from the CLI:

python -m tf2onnx.convert --saved-model $tf_model_name --output $onnx_model_name

In [12]:

import logging
import runpy
import sys

old_argv = sys.argv

try:
    sys.argv = ["python", "--saved-model", tf_model_name, "--output", onnx_model_name]
    runpy.run_module("tf2onnx.convert", run_name="__main__")
finally:
    sys.argv = old_argv
    logging.getLogger().setLevel(logging.WARNING)

Out[12]:

<frozen runpy>:128: RuntimeWarning: 'tf2onnx.convert' found in sys.modules after import of package 'tf2onnx', but prior to execution of 'tf2onnx.convert'; this may result in unpredictable behaviour
2026-02-16 08:12:33,049 - WARNING - ***IMPORTANT*** Installed protobuf is not cpp accelerated. Conversion will be extremely slow. See https://github.com/onnx/tensorflow-onnx/issues/1557
2026-02-16 08:12:33,051 - WARNING - '--tag' not specified for saved_model. Using --tag serve
2026-02-16 08:12:33,190 - INFO - Signatures found in model: [serving_default].
2026-02-16 08:12:33,191 - WARNING - '--signature_def' not specified, using first signature: serving_default
2026-02-16 08:12:33,191 - INFO - Output names: ['output_0']
2026-02-16 08:12:33,191 - WARNING - Could not search for non-variable resources. Concrete function internal representation may have changed.
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1771225953.195756   41231 devices.cc:76] Number of eligible GPUs (core count >= 8, compute capability >= 0.0): 0 (Note: TensorFlow was not compiled with CUDA or ROCm support)
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1771225953.195882   41231 single_machine.cc:361] Starting new session
I0000 00:00:1771225953.242144   41231 devices.cc:76] Number of eligible GPUs (core count >= 8, compute capability >= 0.0): 0 (Note: TensorFlow was not compiled with CUDA or ROCm support)
I0000 00:00:1771225953.242191   41231 single_machine.cc:361] Starting new session
2026-02-16 08:12:33,260 - INFO - Using tensorflow=2.18.0, onnx=1.17.0, tf2onnx=1.16.0/5dd776
2026-02-16 08:12:33,261 - INFO - Using opset <onnx, 15>
2026-02-16 08:12:33,266 - INFO - Computed 0 values for constant folding
2026-02-16 08:12:33,273 - INFO - Optimizing ONNX model
2026-02-16 08:12:33,293 - INFO - After optimization: Cast -1 (1->0), Const +1 (9->10), Identity -2 (2->0), Reshape +1 (1->2), Transpose -5 (6->1)
2026-02-16 08:12:33,295 - INFO - 
2026-02-16 08:12:33,296 - INFO - Successfully converted TensorFlow model mnist-tf-model to ONNX
2026-02-16 08:12:33,296 - INFO - Model inputs: ['inputs']
2026-02-16 08:12:33,296 - INFO - Model outputs: ['output_0']
2026-02-16 08:12:33,296 - INFO - ONNX model is saved at tf_model.onnx

In [13]:

import onnx

onnx_model = onnx.load(onnx_model_name)

3.2. (OPTIONAL) Converting TensorFLow Keras model

To convert the TensorFLow Keras model to ONNX format, change the Markdown cells below to a Code cells and remove the triple backticks (```).

!tar xzf MNIST_DATA/{model_tar_name}

import logging

from keras.models import load_model

logging.getLogger().setLevel(logging.ERROR)

model = load_model(tf_model_name)
onnx_model, _ = tf2onnx.convert.from_keras(model)
onnx.save(onnx_model, onnx_model_name)

3.3. Evaluating the ONNX Model

After exporting the model, you should verify its integrity and ensure that it functions as expected. We will use onnxruntime to load the model and perform inference on the test data. Additionally, we’ll use onnx's checker module to validate the exported ONNX model.

In [14]:

onnx.checker.check_model(onnx_model)

In [15]:

import onnxruntime as ort

session = ort.InferenceSession(onnx_model_name)
x_test = np.array(x_test, dtype=np.float32).reshape(-1, 28, 28, 1)
input_data = {session.get_inputs()[0].name: x_test[:2]}
session.run([], input_data)

Out[15]:

[array([[2.9459990e-11, 2.6513208e-10, 5.7859739e-09, 1.4021280e-08,
         9.0819036e-10, 2.0454895e-07, 3.2107092e-08, 9.9999964e-01,
         8.8907576e-08, 3.6024445e-10],
        [1.1899501e-06, 2.6128285e-03, 9.7718209e-01, 1.6054348e-06,
         4.1321553e-07, 1.1023610e-06, 2.0009710e-02, 3.6875292e-08,
         1.9093615e-04, 1.7123283e-07]], dtype=float32)]

4. Persisting converted ONNX model

In this section, you will learn how to store your converted ONNX model in watsonx.ai Runtime repository using the IBM watsonx.ai SDK.

4.1. Publishing model in watsonx.ai Runtime repository

Define model name, type and software spec.

In [16]:

software_spec_id = client.software_specifications.get_id_by_name("onnxruntime_opset_19")
onnx_model_zip = "tf_onnx.zip"

In [17]:

import zipfile

with zipfile.ZipFile(onnx_model_zip, "w", zipfile.ZIP_DEFLATED) as zip_file:
    zip_file.write(onnx_model_name)

In [18]:

metadata = {
    client.repository.ModelMetaNames.NAME: "TensorFlow to ONNX converted model",
    client.repository.ModelMetaNames.TYPE: "onnxruntime_1.16",
    client.repository.ModelMetaNames.SOFTWARE_SPEC_ID: software_spec_id,
}

published_model = client.repository.store_model(
    model=onnx_model_zip, meta_props=metadata
)

4.2. Getting model details

In [19]:

import json

published_model_id = client.repository.get_model_id(published_model)
model_details = client.repository.get_details(published_model_id)
print(json.dumps(model_details, indent=2))

Out[19]:

{
  "metadata": {
    "name": "TensorFlow to ONNX converted model",
    "space_id": "fb3d528a-bf16-460e-bcf6-06f05d8ba57c",
    "resource_key": "3a8eb861-6f55-4dff-851d-eac2c117dfe7",
    "id": "2c00ba66-2fac-4983-9a01-36bb5338db62",
    "created_at": "2026-02-16T07:13:09Z",
    "rov": {
      "member_roles": {
        "IBMid-696000GJGB": {
          "user_iam_id": "IBMid-696000GJGB",
          "roles": [
            "OWNER"
          ]
        }
      }
    },
    "owner": "IBMid-696000GJGB"
  },
  "entity": {
    "software_spec": {
      "id": "368d2795-aaa7-59a0-834c-248c64a5a99e"
    },
    "type": "onnxruntime_1.16"
  }
}

5. Deploying and scoring ONNX model

In this section you'll learn how to create an online scoring service and predict on unseen data.

5.1. Creating online deployment for published model

In [20]:

metadata = {
    client.deployments.ConfigurationMetaNames.NAME: "Deployment of TensorFLow to ONNX converted model",
    client.deployments.ConfigurationMetaNames.ONLINE: {},
}

created_deployment = client.deployments.create(published_model_id, meta_props=metadata)

Out[20]:

######################################################################################

Synchronous deployment creation for id: '2c00ba66-2fac-4983-9a01-36bb5338db62' started

######################################################################################


initializing
Note: online_url and serving_urls are deprecated and will be removed in a future release. Use inference instead.
......
ready


-----------------------------------------------------------------------------------------------
Successfully finished deployment creation, deployment_id='10d18703-6032-41ed-87aa-53d71abd101f'
-----------------------------------------------------------------------------------------------

In [21]:

deployment_id = client.deployments.get_id(created_deployment)

Now you can print an online scoring endpoint.

In [22]:

client.deployments.get_scoring_href(created_deployment)

5.2. Getting deployment details

In [23]:

client.deployments.get_details(deployment_id)

5.3. Scoring

You can use below method to do test scoring request against deployed model.

Let's first visualize two samples from dataset, we'll use for scoring.

In [24]:

%matplotlib inline

In [25]:

import matplotlib.pyplot as plt

for i, image in enumerate([x_test[0], x_test[1]]):
    plt.subplot(2, 2, i + 1)
    plt.axis("off")
    plt.imshow(image.reshape((28, 28)), cmap=plt.cm.gray_r, interpolation="nearest")

Out[25]:

Prepare scoring payload with records to score.

In [26]:

scoring_payload = {"input_data": [{"values": x_test[:2].tolist()}]}

Use client.deployments.score() method to run scoring.

In [27]:

predictions = client.deployments.score(deployment_id, scoring_payload)

Let's print the result of predictions.

In [28]:

print(json.dumps(predictions, indent=2))

Out[28]:

{
  "predictions": [
    {
      "id": "output_0",
      "values": [
        [
          2.9460045514184685e-11,
          2.651320829549775e-10,
          5.78599612666153e-09,
          1.402130678229696e-08,
          9.081955743006631e-10,
          2.0454953642001783e-07,
          3.2107216441090713e-08,
          0.9999996423721313,
          8.890791747262483e-08,
          3.6024583316418557e-10
        ],
        [
          1.1899520586666767e-06,
          0.0026128317695111036,
          0.9771818518638611,
          1.6054360685302527e-06,
          4.13216241668124e-07,
          1.102363967220299e-06,
          0.020009761676192284,
          3.687535610197301e-08,
          0.00019093610171694309,
          1.7123278439612477e-07
        ]
      ]
    }
  ]
}

In [29]:

for i, prediction in enumerate(predictions["predictions"][0]["values"]):
    plt.bar(range(10), prediction)
    plt.title("Digit probability")
    plt.xlabel("Digit")
    plt.ylabel("Probability")
    plt.xticks(range(10))
    plt.yscale("log")
    plt.show()

Out[29]:

As you can see, the predicted values are consistent with those calculated in the evaluation section.

6. Cleaning up

If you want to clean up after the notebook execution, i.e. remove any created assets like:

experiments
trainings
pipelines
model definitions
models
functions
deployments

please follow up this sample notebook.

7. Summary and next steps

You successfully completed this notebook! You learned how to use ONNX, TensorFlow machine learning library as well as watsonx.ai for model creation and deployment. Check out our Online Documentation for more samples, tutorials, documentation, how-tos, and blog posts.

Authors

Michał Koruszowic, Software Engineer

Use ONNX model converted from TensorFlow to recognize hand-written digits with `ibm-watsonx-ai`

Learning goals

Contents

1. Setting up the environment

Install dependencies

Connection to watsonx.ai Runtime

1.3. Working with spaces

2. Downloading externally created TensorFlow model and data

2.1. Downloading dataset

2.2. Downloading TensorFlow model

3. Converting TensorFlow model to ONNX format

3.1. Converting TensorFlow SavedModel

3.2. (OPTIONAL) Converting TensorFLow Keras model

3.3. Evaluating the ONNX Model

4. Persisting converted ONNX model

4.1. Publishing model in watsonx.ai Runtime repository

4.2. Getting model details

5. Deploying and scoring ONNX model

5.1. Creating online deployment for published model

5.2. Getting deployment details

5.3. Scoring

6. Cleaning up

7. Summary and next steps

Authors

Product

Resources

Company

Use ONNX model converted from TensorFlow to recognize hand-written digits with ibm-watsonx-ai

Learning goals

Contents

1. Setting up the environment

Install dependencies

Connection to watsonx.ai Runtime

1.3. Working with spaces

2. Downloading externally created TensorFlow model and data

2.1. Downloading dataset

2.2. Downloading TensorFlow model

3. Converting TensorFlow model to ONNX format

3.1. Converting TensorFlow SavedModel

3.2. (OPTIONAL) Converting TensorFLow Keras model

3.3. Evaluating the ONNX Model

4. Persisting converted ONNX model

4.1. Publishing model in watsonx.ai Runtime repository

4.2. Getting model details

5. Deploying and scoring ONNX model

5.1. Creating online deployment for published model

5.2. Getting deployment details

5.3. Scoring

6. Cleaning up

7. Summary and next steps

Authors

Use ONNX model converted from TensorFlow to recognize hand-written digits with `ibm-watsonx-ai`