Easy Zero-Knowledge Inference

ezkl is a library and command-line tool for doing inference for deep learning models and other computational graphs in a zk-snark (ZKML). It enables the following workflow:

1. Define a computational graph, for instance a neural network (but really any arbitrary set of operations), as you would normally in pytorch or tensorflow.
2. Export the final graph of operations as an .onnx file and some sample inputs to a .json file.
3. Point ezkl to the .onnx and .json files to generate a ZK-SNARK circuit with which you can prove statements such as:

"I ran this publicly available neural network on some private data and it produced this output"

"I ran my private neural network on some public data and it produced this output"

"I correctly ran this publicly available neural network on some public data and it produced this output"

In the backend we use Halo2 as a proof system.

The generated proofs can then be used on-chain to verify computation, only the Ethereum Virtual Machine (EVM) is supported at the moment.

• If you have any questions, we'd love for you to open up a discussion topic in Discussions. Alternatively, you can join the ✨EZKL Community Telegram Group💫.

• For more technical writeups and details check out our blog.

• To see what you can build with ezkl, check out cryptoidol.tech where ezkl is used to create an AI that judges your singing ... forever.

Install the python bindings by calling.

pip install ezkl

Or for the GPU:

pip install ezkl-gpu

Google Colab Example to learn how you can train a neural net and deploy an inference verifier onchain for use in other smart contracts.

More notebook tutorials can be found within examples/notebooks.

Install the CLI

curl https://hub.ezkl.xyz/install_ezkl_cli.sh | bash

For more details visit the docs.

Build the auto-generated rust documentation and open the docs in your browser locally. cargo doc --open

You can install the library from source

cargo install --locked --path .

You will need a functioning installation of solc in order to run ezkl properly. solc-select is recommended. Follow the instructions on solc-select to activate solc in your environment.

Python bindings exists and can be built using maturin. You will need rust and cargo to be installed.

python -m venv .env
source .env/bin/activate
pip install -r requirements.txt
maturin develop --release --features python-bindings
# dependencies specific to tutorials
pip install torch pandas numpy seaborn jupyter onnx kaggle py-solc-x web3 librosa tensorflow keras tf2onnx

If you have access to NVIDIA GPUs, you can enable acceleration by building with the feature icicle and setting the following environment variable:

export ENABLE_ICICLE_GPU=true

GPU acceleration is provided by Icicle

To go back to running with CPU, the previous environment variable must be unset instead of being switch to a value of false:

unset ENABLE_ICICLE_GPU

NOTE: Even with the above environment variable set, icicle is disabled for circuits where k <= 8. To change the value of k where icicle is enabled, you can set the environment variable ICICLE_SMALL_K.

The EZKL project has several libraries and repos.

If you're interested in contributing and are unsure where to start, reach out to one of the maintainers:

• dante (alexander-camuto)
• jason (jasonmorton)

More broadly:

• See currently open issues for ideas on how to contribute.

• For PRs we use the conventional commits naming convention.

• To report bugs or request new features create a new issue within Issues to inform the greater community.

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you shall be licensed to Zkonduit Inc. under the terms and conditions specified in the CLA, which you agree to by intentionally submitting a contribution. In particular, you have the right to submit the contribution and we can distribute it under the Apache 2.0 license, among other terms and conditions.

Ezkl is unaudited, beta software undergoing rapid development. There may be bugs. No guarantees of security are made and it should not be relied on in production.

NOTE: Because operations are quantized when they are converted from an onnx file to a zk-circuit, outputs in python and ezkl may differ slightly.