View source for How to adjust a Neural Networks metadata

Metadata of a Neural Network stored in the [[ONNX]] file format can be adjusted using the python code found below.

==Prerequisites==
An installed [https://en.wikipedia.org/wiki/Python_(programming_language) python] environment is required to execute the code. 
Additionally, install the onnx library using [https://en.wikipedia.org/wiki/Pip_%28package_manager%29 PIP]:
<pre>pip install onnx </pre>

To add or edit a metadata key (without duplication), you can use the following code:
<pre>
def addMeta(onnx_model, key, value):
    
    for entry in onnx_model.metadata_props:
        if entry.key == key:
            entry.value = value
            return

    meta = onnx_model.metadata_props.add()
    meta.key = key
    meta.value = value
</pre>

To adjust an existing ONNX file, you first have to load the model. Next, pass the model to the method, with additional key and value arguments.
<pre>
import onnx
onnx_model = onnx.load("path/to/onnx/file")
addMeta(onnxModel, "key", "value")
</pre>

Lastly, save the onnx model again to a (new) file: 
<pre>
with open(onnxFilePath, "wb") as f:
    f.write(onnx_model.SerializeToString())
</pre>

Producer, version and description are set using these functions, as these are not stored in the meta data.
<pre>
def setProducer(onnxModel, producer):
    onnxModel.producer_name = producer

def setModelVersion(onnxModel, modelVersion):
    onnxModel.model_version = modelVersion

def setDocString(onnxModel, description):
    onnxModel.doc_string = description
</pre>

{{article end
|seealso=*[[Neural Network]]
*[[ONNX]]
}}
{{InferenceOverlay nav}}