ONNX

ONNX, or Open Neural Network Exchange, is an open-source format designed to represent machine learning models. It provides a standardized way to describe models so that they can be easily exchanged between different deep learning frameworks. ONNX is supported by various frameworks such as PyTorch, TensorFlow, Microsoft Cognitive Toolkit (CNTK), and others, allowing interoperability and flexibility in deploying models across different platforms.

Dozer supports ONNX models and allows you to deploy them as APIs. This enables you to use your models in production without having to write any additional code. For instance, you can use a pre-trained model to predict probabilities of a particular event, such as a customer credit score, or other use cases.

Configuration

Add the following block to your YAML file to register ONNX models.

sql : |
    SELECT torch_jit(col1, col2) INTO output FROM input;

udfs:
    - name: torch_jit
      config: !Onnx
        path: ./model.onnx

torch_jit is the function which would run the ONNX model on col1, col2 as input returning the output in output` column.

Parameters

Parameter Name	Type	Description
`path`	String	Path to the ONNX model.

Running the ONNX Model

Pre-requisites

Enabling ONNX feature while building Dozer.

cargo install --path dozer-cli --features onnx --locked

Installing ONNX runtime.
```
  pip install onnxruntime
```

Run App to start ingesting data into Dozer.

dozer run app

To use ONNX on Dozer Cloud

dozer cloud deploy -c dozer-config.yaml -c model.onnx

Trying it out

To test a ONNX sample, clone the dozer-samples GitHub repo and follow the steps described here

ONNX

Configuration​

Parameters​

Running the ONNX Model​

Pre-requisites​

To use ONNX on Dozer Cloud​

Trying it out​