Skip to content

Tutorial 02 — Full zk-SNARK Run

Goal: Generate real zero-knowledge proofs for inclusion and exclusion. This requires a one-time circuit setup (compilation + key generation) and then proof generation.

Prerequisites: - Python dependencies installed (pip install -r requirements.txt) - EZKL installed and functional (verify with python -c "import ezkl; print(ezkl.__version__)") - Tutorial 01 completed (understanding of the dry-run output)

Overview of what happens

``` Setup phase (once per circuit configuration) └── Compile PyTorch model → ONNX → arithmetic circuit └── Generate proving key (pk) and verification key (vk) └── Write artifacts to proofs///

Proof phase (per user record) └── Run EZKL prove on the compiled circuit └── Write .pf proof file └── Verify .pf file against vk ```

Step 1 — Run setup and proof generation together

The following command runs everything from scratch:

bash ZKP_MODE=1 SETUP_MRP=1 SETUP_LTR=1 GEN_FULL_PROOF=1 python driver.py

Variable Value Effect
ZKP_MODE 1 Enables actual proof generation
SETUP_MRP 1 Compiles the Merkle path aggregator circuit
SETUP_LTR 1 Compiles the leaf transform circuit
GEN_FULL_PROOF 1 Generates a proof at every level of the SMT path

Time warning: Setup involves EZKL circuit compilation and can take several minutes per circuit on a laptop, depending on model complexity and TREE_HEIGHT.

Step 2 — Note the generated UUID

When setup completes, a UUID is printed (or logged) and artifacts are written to:

proofs/simple_sum_flow-through/<uuid>/ ├── ltr_keys/ │ ├── network.onnx │ ├── model.compiled │ ├── pk.key │ ├── vk.key │ └── settings.json └── mrp_keys/ ├── network.onnx ├── model.compiled ├── pk.key ├── vk.key └── settings.json

Copy and save the UUID — you will need it to skip setup on future runs.

Step 3 — Reuse circuits on subsequent runs

Pass the saved UUID as ID to skip the expensive setup phase:

bash ID=<your-uuid> ZKP_MODE=1 GEN_FULL_PROOF=1 python driver.py

Proof generation still runs, but circuit compilation is skipped. This is much faster.

Step 4 — Examine the proof files

After a successful run, .pf files are written alongside the keys. You can verify any proof file manually:

```bash python - <<'EOF' import ezkl, asyncio

async def check(): ok = await ezkl.verify( proof_path="proofs/simple_sum_flow-through//ltr_keys/test_.pf", settings_path="proofs/simple_sum_flow-through//ltr_keys/settings.json", vk_path="proofs/simple_sum_flow-through//ltr_keys/vk.key", ) print("Proof valid:", ok)

asyncio.run(check()) EOF ```

Step 5 — Faster variant (skip default-value levels)

Setting GEN_FULL_PROOF=0 generates proofs only at SMT levels where a sibling is non-default (i.e., another real user shares a path prefix). This is much faster and still proves membership:

bash ID=<your-uuid> ZKP_MODE=1 GEN_FULL_PROOF=0 python driver.py

Troubleshooting

"Dimension mismatch" or ONNX shape error

Inspect the compiled ONNX graph:

bash python -c ' import onnx m = onnx.load("proofs/simple_sum_flow-through/<uuid>/ltr_keys/network.onnx") print(onnx.helper.printable_graph(m.graph)) '

Common fix: ensure transform_salt_shape in driver.py matches the actual salt tensor dimensions.

"ID not found" error

The UUID you passed does not have a matching subdirectory under proofs/. Either: - Leave ID unset to generate a new one. - Check the path proofs/<prover_name>/<uuid>/ exists.

Out of memory during setup

Reduce TREE_HEIGHT or use GEN_FULL_PROOF=0. Smaller trees require smaller circuits.

What's next