How vt-asaplab/zip works
ZIP is a research prototype for high-precision zero-knowledge AI inference. The documented workflow combines neural-network inference workloads with precomputed piecewise-polynomial approximations and proof-generation components: linear PLONK circuits use zk-Location, while nonlinear circuits and a multi-lookup argument are part of ZIP. The graph also shows NFGen’s documented generator dependencies and Caulk’s single- and multi-opening code. Several subsystem links come from the README; unsampled source is represented at directory level where implementation details are uncertain.
Main entry points: main.py NFGen generator · multi/unity.rs Multi lookup · approx.py Approximation driver · single/unity.rs Single lookup
Components 19
Inference workloads
- LeNet workload
src/CNN/LeNet/infer_lenet.py - UTKFace CNN
src/CNN/UTKFace_MAE/CNN_MAE.py - miniBERT workload
src/LLM/miniBERT/minibert.py
Polynomial approximation
- Approximation driver
src/piecewise_polynomial_approximation/approx.py - NFGen generator
src/piecewise_polynomial_approximation/NFGen/src/NFGen/main.py - Polynomial fitting
src/piecewise_polynomial_approximation/NFGen/src/NFGen/fitter.py - Fit analysis
src/piecewise_polynomial_approximation/NFGen/src/NFGen/analysis.py - Function utilities
src/piecewise_polynomial_approximation/NFGen/src/NFGen/funcs.py - Code generation
src/piecewise_polynomial_approximation/NFGen/src/NFGen/code_generator.py - Activation tables
src/piecewise_polynomial_approximation/precomputed_lookup_tables_ieee754/elu_approx.py
Proof generation
- Linear circuits
src/proof_generation/ZIP_proof_generation/ZIP_circuit/circuit/linear/main.go - Nonlinear circuits
src/proof_generation/ZIP_proof_generation/ZIP_circuit/circuit/non-linear/main.go - ZIP multi-lookup
src/proof_generation/ZIP_proof_generation/ZIP_lookup/add_interval_index.py - IEEE linear proofs
src/proof_generation/zk-Location
Proof protocols
- Single lookup
src/proof_generation/caulk/src/single/unity.rs - Multi lookup
src/proof_generation/caulk/src/multi/unity.rs - Lookup setup
src/proof_generation/caulk/src/multi/setup.rs - KZG backend
src/proof_generation/caulk/src/kzg.rs
Other components
- Research user
Connections 19
- Research user → LeNet workload: runs inference
README.md - Research user → UTKFace CNN: runs inference
README.md - Research user → miniBERT workload: runs inference
README.md - Approximation driver → NFGen generator: invokes generator
src/piecewise_polynomial_approximation/approx.py - NFGen generator → Polynomial fitting: uses fitting
src/piecewise_polynomial_approximation/NFGen/src/NFGen/main.py - NFGen generator → Function utilities: uses functions
src/piecewise_polynomial_approximation/NFGen/src/NFGen/main.py - NFGen generator → Fit analysis: uses analysis
src/piecewise_polynomial_approximation/NFGen/src/NFGen/main.py - NFGen generator → Code generation: uses generator
src/piecewise_polynomial_approximation/NFGen/src/NFGen/main.py - UTKFace CNN → Activation tables: imports approximation
src/CNN/UTKFace_MAE/CNN_MAE.py - Approximation driver → Activation tables: generates approximations
src/piecewise_polynomial_approximation/approx.py - Research user → Linear circuits: proves inference
README.md - Research user → Nonlinear circuits: proves inference
README.md - Linear circuits → IEEE linear proofs: uses linear proofs
README.md - Nonlinear circuits → ZIP multi-lookup: uses lookup argument
README.md - ZIP multi-lookup → Multi lookup: integrates multi-lookup
README.md - Multi lookup → Lookup setup: uses setup
src/proof_generation/caulk/src/multi/mod.rs - Multi lookup → KZG backend: uses commitments
src/proof_generation/caulk/src/multi/mod.rs - Single lookup → Lookup setup: uses setup
- Single lookup → KZG backend: uses commitments
src/proof_generation/caulk/src/single/unity.rs