# Setup

The lifecycle of an EZKL proof consists of three core components: Setup, Prove, and Verify. This page focuses on the Setup phase, which defines the rules for proof generation and verification. Recall that:

Setup: Defines proof parameters and generates keys (performed by developers)
Prove: Generates a proof based on the setup (performed by users)
Verify: Checks the validity of a proof (performed by verifiers)

# Process

The setup process involves the following steps:

Generate settings: Creates a configuration file with default parameters for the circuit.
Calibrate settings (optional): Fine-tunes the circuit parameters to optimize for either accuracy or resource usage.
Compile the model: Converts the ONNX model into a format optimized for zero-knowledge proofs.
Run setup: Generates the cryptographic keys needed for proving and verifying.

# Parameters

To perform the setup, you'll need to provide and/or create the following:

# ONNX File

For PyTorch:

import torch.onnx

dummy_input = torch.randn(1, 3, 224, 224)  # Adjust input dimensions as needed
torch.onnx.export(model, dummy_input, "network.onnx", opset_version=10)

For TensorFlow:

import tf2onnx

onnx_model, _ = tf2onnx.convert.from_keras(model)
with open("network.onnx", "wb") as f:
    f.write(onnx_model.SerializeToString())

# Structured Reference String (SRS)

The SRS is a common, public piece of cryptographic data. You can download it using the EZKL CLI:

ezkl get-srs

# Configuration Options

These are defined in the settings.json file, which is generated and optionally calibrated during the setup process.

# Instructions

Generate settings:

ezkl gen-settings

This creates a settings.json file with default parameters based on your ONNX model.

Calibrate settings (optional):

ezkl calibrate-settings

This optimizes the settings for resource usage. You can also use --target accuracy to optimize for accuracy, and --target resources to optimize for resources.

Compile model:

ezkl compile-circuit

This converts your ONNX model into an optimized format for zero-knowledge proofs, creating the network.ezkl file.

Run setup:

ezkl setup

This generates the cryptographic keys needed for proving and verifying.

Generate settings:

import ezkl

ezkl.gen_settings("network.onnx")

This creates a settings.json file with default parameters based on your ONNX model.

Calibrate settings (optional):

ezkl.calibrate_settings("network.onnx", "settings.json", target="resources")

This optimizes the settings for resource usage. You can also use target="accuracy" to optimize for accuracy.

Compile model:

ezkl.compile_circuit("network.onnx", "network.ezkl", "settings.json")

This converts your ONNX model into an optimized format for zero-knowledge proofs, creating the network.ezkl file.

Run setup:

ezkl.setup("network.ezkl", "vk.key", "pk.key", "kzg.srs")

This generates the cryptographic keys needed for proving and verifying.

Generate settings:

import { Engine } from '@ezkljs/engine';

const engine = new Engine();
await engine.genSettings("network.onnx");

This creates a settings.json file with default parameters based on your ONNX model.

Calibrate settings (optional):

await engine.calibrateSettings("network.onnx", "settings.json", { target: "resources" });

This optimizes the settings for resource usage. You can also use { target: "accuracy" } to optimize for accuracy.

Compile model:

await engine.compileCircuit("network.onnx", "network.ezkl", "settings.json");

This converts your ONNX model into an optimized format for zero-knowledge proofs, creating the network.ezkl file.

Run setup:

await engine.setup("network.ezkl", "vk.key", "pk.key", "kzg.srs");

This generates the cryptographic keys needed for proving and verifying.

Upload your files,

archon create-artifact -a test -i input.json -m network.onnx -c calibration.json

Generate settings:

archon job -a test gen-settings

This creates a settings.json file with default parameters based on your ONNX model.

Calibrate settings (optional):

archon job -a test calibrate-settings

This optimizes the settings for resource usage. You can also use --target accuracy to optimize for accuracy, and --target resources to optimize for resources.

Compile model:

archon job -a test compile-circuit

This converts your ONNX model into an optimized format for zero-knowledge proofs, creating the network.ezkl file.

Run setup:

archon job -a test setup

This generates the cryptographic keys needed for proving and verifying.

# Outputs

The setup process will generate the following:

network.onnx: model in ONNX format
settings.json: generated file which is optionally calibrated during the setup process
network.ezkl: compiled circuit created from your ONNX model in step 3.
kzg.srs: Structured Reference String (SRS), which you can download using ezkl get-srs.
vk.key and pk.key: verification and proving keys generated during the setup process