[Step 4] Model Training

Please refer to our Starter Kit to see how you can participate as a Modeler!

Hard Requirements: ML Models

• Supported models: neural network using PyTorch (other than recurrent neural networks, e.g., LSTM) and logistic regression using PyTorch

• A model's output directly has to be the required target label

  • For Challenge 1, this means the model's output must be a probability (as opposed to a logit)

• A model must be exported to the ONNX format

Soft Requirements: Compatibility with zkML

We encourage Modelers to try out various PyTorch NN architectures and also to run the end-to-end zkML workflow on their local machines to ensure that their submitted models:

• can be exported to the ONNX format

• have minimum quantization error

• are practical in terms of time required to generate a single zkML proof and compute, disk, and memory requirements (note that the model's complexity directly correlates with the zkML overhead)