MCPcopy Index your code
hub / github.com/ddkang/zkml

github.com/ddkang/zkml @main

Chat with this repo
repository ↗ · DeepWiki ↗ · + Follow
492 symbols 855 edges 87 files 18 documented · 4%
What it actually does AI analysis from the code graph — generated when you open this
loading…
README

zkml

zkml is a framework for constructing proofs of ML model execution in ZK-SNARKs. Read our blog post and paper for implementation details.

zkml requires the nightly build of Rust:

rustup override set nightly

Quickstart

Run the following commands:

# Installs rust, skip if you already have rust installed
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh

git clone https://github.com/ddkang/zkml.git
cd zkml
rustup override set nightly
cargo build --release
mkdir params_kzg
mkdir params_ipa

# This should take ~16s to run the first time
# and ~8s to run the second time
./target/release/time_circuit examples/mnist/model.msgpack examples/mnist/inp.msgpack kzg

This will prove an MNIST circuit! It will require around 2GB of memory and take around 8 seconds to run.

Converting your own model and data

To convert your own model and data, you will need to convert the model and data to the format zkml expects. Currently, we accept TFLite models. We show an example below.

  1. First cd examples/mnist

  2. We've already created a model that achieves high accuracy on MNIST (model.tflite). You will need to create your own TFLite model. One way is to convert a model from Keras.

  3. You will need to convert the model:

python ../../python/converter.py --model model.tflite --model_output converted_model.msgpack --config_output config.msgpack --scale_factor 512 --k 17 --num_cols 10 --num_randoms 1024

There are several parameters that need to be changed depending on the model (scale_factor, k, num_cols, and num_randoms).

  1. You will first need to serialize the model input to numpy's serialization format npy. We've written a small script to do this for the first test data point in MNIST:
python data_to_npy.py
  1. You will then need to convert the input to the model:
python ../../python/input_converter.py --model_config converted_model.msgpack --inputs 7.npy --output example_inp.msgpack
  1. Once you've converted the model and input, you can run the model as above! However, we generally recommend testing the model before proving (you will need to build zkml before running the next line):
cd ../../
./target/release/test_circuit examples/mnist/converted_model.msgpack examples/mnist/example_inp.msgpack

Contact us

If you're interested in extending or using zkml, please contact us at ddkang [at] g.illinois.edu.

Extension points exported contracts — how you extend this code

Layer (Interface)
General issue with rust: I'm not sure how to pass named arguments to a trait... Currently, the caller must be aware of t [33 …
src/layers/layer.rs
Gadget (Interface)
(no doc) [23 implementers]
src/gadgets/gadget.rs
Commit (Interface)
(no doc) [1 implementers]
src/commitments/commit.rs
GadgetConsumer (Interface)
(no doc) [34 implementers]
src/layers/layer.rs
NonLinearGadget (Interface)
(no doc) [7 implementers]
src/gadgets/nonlinear/non_linearity.rs
Arithmetic (Interface)
(no doc) [4 implementers]
src/layers/arithmetic.rs
Averager (Interface)
(no doc) [2 implementers]
src/layers/averager.rs

Core symbols most depended-on inside this repo

shape
called by 80
src/layers/max_pool_2d.rs
op_aligned_rows
called by 18
src/gadgets/gadget.rs
load_lookups
called by 13
src/gadgets/input_lookup.rs
convert_to_u128
called by 10
src/gadgets/gadget.rs
convert_to_u64
called by 9
src/gadgets/gadget.rs
new_tensor
called by 9
backwards/backward.py
num_inputs_per_row
called by 7
src/gadgets/max.rs
forward
called by 6
src/layers/softmax.rs

Shape

Method 368
Class 80
Function 32
Interface 7
Enum 5

Languages

Rust93%
Python7%

Modules by API surface

backwards/backward.py19 symbols
src/model.rs14 symbols
src/gadgets/nonlinear/tanh.rs13 symbols
src/gadgets/nonlinear/sqrt.rs13 symbols
src/gadgets/nonlinear/rsqrt.rs13 symbols
src/gadgets/nonlinear/relu.rs13 symbols
src/gadgets/nonlinear/pow.rs13 symbols
src/gadgets/nonlinear/logistic.rs13 symbols
src/gadgets/nonlinear/exp.rs13 symbols
src/layers/conv2d.rs11 symbols
src/gadgets/dot_prod.rs11 symbols
src/gadgets/bias_div_round_relu6.rs11 symbols

For agents

$ claude mcp add zkml \
  -- python -m otcore.mcp_server <graph>

⬇ download graph artifact