refactor: selectable key ser (#718)

.github/workflows/large-tests.yml

@@ -11,7 +11,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - name: nanoGPT Mock

.github/workflows/release.yml

@@ -45,7 +45,7 @@ jobs:
     steps:
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - name: Checkout repo

.github/workflows/rust.yml

@@ -26,7 +26,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - name: Build
@@ -38,7 +38,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - name: Docs
@@ -50,7 +50,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -73,7 +73,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -106,7 +106,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -139,7 +139,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -172,7 +172,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -189,7 +189,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: jetli/wasm-pack-action@v0.4.0
@@ -199,7 +199,7 @@ jobs:
       - name: Install wasm32-unknown-unknown
         run: rustup target add wasm32-unknown-unknown
       - name: Add rust-src
-        run: rustup component add rust-src --toolchain nightly-2023-08-24-x86_64-unknown-linux-gnu
+        run: rustup component add rust-src --toolchain nightly-2024-01-04-x86_64-unknown-linux-gnu
       - name: Run wasm verifier tests
         # on mac:
         #   AR=/opt/homebrew/opt/llvm/bin/llvm-ar CC=/opt/homebrew/opt/llvm/bin/clang wasm-pack test --firefox --headless -- -Z build-std="panic_abort,std" --features web
@@ -212,7 +212,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -229,7 +229,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -284,7 +284,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -310,7 +310,7 @@ jobs:
       - name: Install solc
         run: (hash svm 2>/dev/null || cargo install svm-rs) && svm install 0.8.20 && solc --version
       - name: Install Anvil
-        run: cargo install --git https://github.com/foundry-rs/foundry --rev 95a93cd397f25f3f8d49d2851eb52bc2d52dd983 --profile local --locked anvil --force
+        run: cargo install --git https://github.com/foundry-rs/foundry --rev b320f350156a0fb15c2eb13dc380deb2367c4474 --profile local --locked anvil --force
       - name: KZG prove and verify tests (EVM + VK rendered seperately)
         run: cargo nextest run --release --verbose tests_evm::kzg_evm_prove_and_verify_render_seperately_ --test-threads 1
       - name: KZG prove and verify tests (EVM + kzg all)
@@ -343,7 +343,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: jetli/wasm-pack-action@v0.4.0
@@ -351,7 +351,7 @@ jobs:
         run: rustup target add wasm32-unknown-unknown
 
       - name: Add rust-src
-        run: rustup component add rust-src --toolchain nightly-2023-08-24-x86_64-unknown-linux-gnu
+        run: rustup component add rust-src --toolchain nightly-2024-01-04-x86_64-unknown-linux-gnu
       - uses: actions/checkout@v3
       - name: Use pnpm 8
         uses: pnpm/action-setup@v2
@@ -411,11 +411,11 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - name: Add rust-src
-        run: rustup component add rust-src --toolchain nightly-2023-08-24-x86_64-unknown-linux-gnu
+        run: rustup component add rust-src --toolchain nightly-2024-01-04-x86_64-unknown-linux-gnu
       - uses: actions/checkout@v3
       - uses: baptiste0928/cargo-install@v1
         with:
@@ -445,7 +445,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -455,7 +455,7 @@ jobs:
       - name: Install solc
         run: (hash svm 2>/dev/null || cargo install svm-rs) && svm install 0.8.20 && solc --version
       - name: Install Anvil
-        run: cargo install --git https://github.com/foundry-rs/foundry --rev 95a93cd397f25f3f8d49d2851eb52bc2d52dd983 --profile local --locked anvil --force
+        run: cargo install --git https://github.com/foundry-rs/foundry --rev b320f350156a0fb15c2eb13dc380deb2367c4474 --profile local --locked anvil --force
       - name: fuzz tests (EVM)
         run: cargo nextest run --release --verbose tests_evm::kzg_evm_fuzz_ --test-threads 2
       # - name: fuzz tests
@@ -468,7 +468,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -486,7 +486,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -503,7 +503,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -520,7 +520,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -530,7 +530,7 @@ jobs:
       - name: Install solc
         run: (hash svm 2>/dev/null || cargo install svm-rs) && svm install 0.8.20 && solc --version
       - name: Install Anvil
-        run: cargo install --git https://github.com/foundry-rs/foundry --rev 95a93cd397f25f3f8d49d2851eb52bc2d52dd983 --profile local --locked anvil --force
+        run: cargo install --git https://github.com/foundry-rs/foundry --rev b320f350156a0fb15c2eb13dc380deb2367c4474 --profile local --locked anvil --force
       - name: KZG prove and verify aggr tests
         run: cargo nextest run  --release --verbose tests_evm::kzg_evm_aggr_prove_and_verify_::t --test-threads 4 -- --include-ignored
 
@@ -541,7 +541,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -563,7 +563,7 @@ jobs:
           python-version: "3.7"
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - name: Install solc
@@ -571,7 +571,7 @@ jobs:
       - name: Setup Virtual Env and Install python dependencies
         run: python -m venv .env; source .env/bin/activate; pip install -r requirements.txt;
       - name: Install Anvil
-        run: cargo install --git https://github.com/foundry-rs/foundry --rev 95a93cd397f25f3f8d49d2851eb52bc2d52dd983 --profile local --locked anvil --force
+        run: cargo install --git https://github.com/foundry-rs/foundry --rev b320f350156a0fb15c2eb13dc380deb2367c4474 --profile local --locked anvil --force
       - name: Build python ezkl
         run: source .env/bin/activate; maturin develop --features python-bindings --release
       - name: Run pytest
@@ -587,7 +587,7 @@ jobs:
           python-version: "3.7"
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -618,7 +618,7 @@ jobs:
           python-version: "3.9"
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -628,7 +628,7 @@ jobs:
       - name: Install solc
         run: (hash svm 2>/dev/null || cargo install svm-rs) && svm install 0.8.20 && solc --version
       - name: Install Anvil
-        run: cargo install --git https://github.com/foundry-rs/foundry --rev 95a93cd397f25f3f8d49d2851eb52bc2d52dd983 --profile local --locked anvil --force
+        run: cargo install --git https://github.com/foundry-rs/foundry --rev b320f350156a0fb15c2eb13dc380deb2367c4474 --profile local --locked anvil --force
       - name: Setup Virtual Env and Install python dependencies
         run: python -m venv .env; source .env/bin/activate; pip install -r requirements.txt;
       - name: Build python ezkl

.github/workflows/wasm.yml

@@ -22,7 +22,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: jetli/wasm-pack-action@v0.4.0
@@ -30,7 +30,7 @@ jobs:
         run: rustup target add wasm32-unknown-unknown
 
       - name: Add rust-src
-        run: rustup component add rust-src --toolchain nightly-2023-08-24-x86_64-unknown-linux-gnu
+        run: rustup component add rust-src --toolchain nightly-2024-01-04-x86_64-unknown-linux-gnu
       - name: Install binaryen
         run: |
             set -e

Cargo.toml

@@ -17,7 +17,7 @@ crate-type = ["cdylib", "rlib"]
 [dependencies]
 halo2_gadgets = { git = "https://github.com/zkonduit/halo2", branch= "main" }
 halo2_proofs = { git = "https://github.com/zkonduit/halo2", branch= "main"  }
-halo2curves = { version = "0.6.0", features = ["derive_serde"] }
+halo2curves = { git = "https://github.com/privacy-scaling-explorations/halo2curves", rev="9fff22c", features=["derive_serde"] }
 rand = { version = "0.8", default_features = false }
 itertools = { version = "0.10.3", default_features = false }
 clap = { version = "4.3.3", features = ["derive"]}
@@ -34,10 +34,11 @@ bincode = { version = "1.3.3", default_features = false }
 ark-std = { version = "^0.3.0", default-features = false }
 unzip-n = "0.1.2"
 num = "0.4.1"
+portable-atomic = "1.6.0"
 
 # evm related deps
 [target.'cfg(not(target_arch = "wasm32"))'.dependencies]
-ethers = { version = "2.0.7", default_features = false, features = ["ethers-solc"] }
+ethers = { version = "2.0.11", default_features = false, features = ["ethers-solc"] }
 indicatif = {version = "0.17.5", features = ["rayon"]}
 gag =  { version = "1.0.0", default_features = false}
 instant = { version = "0.1" }

examples/conv2d_mnist/main.rs

@@ -6,6 +6,7 @@ use ezkl::fieldutils;
 use ezkl::fieldutils::i32_to_felt;
 use ezkl::tensor::*;
 use halo2_proofs::dev::MockProver;
+use halo2_proofs::poly::commitment::Params;
 use halo2_proofs::poly::kzg::multiopen::{ProverSHPLONK, VerifierSHPLONK};
 use halo2_proofs::{
     circuit::{Layouter, SimpleFloorPlanner, Value},
@@ -489,6 +490,7 @@ pub fn runconv() {
         strategy,
         pi_for_real_prover,
         &mut transcript,
+        params.n(),
     );
     assert!(verify.is_ok());
 

src/circuit/modules/kzg.rs

@@ -41,7 +41,7 @@ pub struct KZGChip {
 }
 
 impl KZGChip {
-    /// Returns the number of inputs to the hash function
+    /// Commit to the message using the KZG commitment scheme
     pub fn commit(
         message: Vec<Fp>,
         degree: u32,

src/circuit/modules/planner.rs

@@ -15,7 +15,7 @@ use halo2_proofs::{
         Instance, Selector, TableColumn,
     },
 };
-use log::{trace, warn};
+use log::{debug, trace};
 
 /// A simple [`FloorPlanner`] that performs minimal optimizations.
 #[derive(Debug)]
@@ -119,7 +119,7 @@ impl<'a, F: Field, CS: Assignment<F> + 'a + SyncDeps> Layouter<F> for ModuleLayo
                 Error::Synthesis
             })?;
             if !self.regions.contains_key(&index) {
-                warn!("spawning module {}", index)
+                debug!("spawning module {}", index)
             };
             self.current_module = index;
         }

src/circuit/ops/hybrid.rs

@@ -6,7 +6,6 @@ use crate::{
     tensor::{self, Tensor, TensorError, TensorType, ValTensor},
 };
 use halo2curves::ff::PrimeField;
-use itertools::Itertools;
 use serde::{Deserialize, Serialize};
 // import run args from model
 
@@ -69,14 +68,6 @@ pub enum HybridOp {
         dim: usize,
         num_classes: usize,
     },
-    GatherElements {
-        dim: usize,
-        constant_idx: Option<Tensor<usize>>,
-    },
-    ScatterElements {
-        dim: usize,
-        constant_idx: Option<Tensor<usize>>,
-    },
 }
 
 impl<F: PrimeField + TensorType + PartialOrd> Op<F> for HybridOp {
@@ -84,7 +75,6 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for HybridOp {
     fn requires_homogenous_input_scales(&self) -> Vec<usize> {
         match self {
             HybridOp::Greater | HybridOp::Less | HybridOp::Equals => vec![0, 1],
-            HybridOp::ScatterElements { .. } => vec![0, 2],
             HybridOp::GreaterEqual | HybridOp::LessEqual => vec![0, 1],
             _ => vec![],
         }
@@ -98,162 +88,42 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for HybridOp {
     fn f(&self, inputs: &[Tensor<F>]) -> Result<ForwardResult<F>, TensorError> {
         let x = inputs[0].clone().map(|x| felt_to_i128(x));
 
-        let (res, intermediate_lookups) = match &self {
-            HybridOp::ReduceMax { axes, .. } => {
-                let res = tensor::ops::max_axes(&x, axes)?;
-                let max_minus_one =
-                    Tensor::from(vec![x.clone().into_iter().max().unwrap() - 1].into_iter());
-                let unit = Tensor::from(vec![1].into_iter());
-                // relu(x - max(x - 1)
-                let inter_1 = (x.clone() - max_minus_one)?;
-                // relu(1 - sum(relu(inter_1)))
-                let inter_2 = (unit
-                    - tensor::ops::sum(&tensor::ops::nonlinearities::leakyrelu(&inter_1, 0.0))?)?;
-
-                (res.clone(), vec![inter_1, inter_2])
-            }
-            HybridOp::ReduceMin { axes, .. } => {
-                let res = tensor::ops::min_axes(&x, axes)?;
-                let min_plus_one =
-                    Tensor::from(vec![x.clone().into_iter().min().unwrap() + 1].into_iter());
-                let unit = Tensor::from(vec![1].into_iter());
-                // relu(min(x + 1) - x)
-                let inter_1 = (min_plus_one - x.clone())?;
-                // relu(1 - sum(relu(inter_1)))
-                let inter_2 = (unit
-                    - tensor::ops::sum(&tensor::ops::nonlinearities::leakyrelu(&inter_1, 0.0))?)?;
-                (res.clone(), vec![inter_1, inter_2])
-            }
-            HybridOp::Div {
-                denom,
-                use_range_check_for_int,
-                ..
-            } => {
-                let res = crate::tensor::ops::nonlinearities::const_div(&x, denom.0 as f64);
-                // if denom is a round number and use_range_check_for_int is true, use range check check
-                if denom.0.fract() == 0.0 && *use_range_check_for_int {
-                    let divisor = Tensor::from(vec![denom.0 as i128 / 2].into_iter());
-                    (res, vec![-divisor.clone(), divisor])
-                } else {
-                    (res, vec![x])
-                }
+        let res = match &self {
+            HybridOp::ReduceMax { axes, .. } => tensor::ops::max_axes(&x, axes)?,
+            HybridOp::ReduceMin { axes, .. } => tensor::ops::min_axes(&x, axes)?,
+            HybridOp::Div { denom, .. } => {
+                crate::tensor::ops::nonlinearities::const_div(&x, denom.0 as f64)
             }
             HybridOp::Recip {
                 input_scale,
                 output_scale,
-                use_range_check_for_int,
-            } => {
-                let res = crate::tensor::ops::nonlinearities::recip(
-                    &x,
-                    input_scale.0 as f64,
-                    output_scale.0 as f64,
-                );
-                // if scale is a round number and use_range_check_for_int is true, use range check check
-                if input_scale.0.fract() == 0.0 && *use_range_check_for_int {
-                    let err_tol = Tensor::from(
-                        vec![(output_scale.0 * input_scale.0) as i128 / 2].into_iter(),
-                    );
-                    (res, vec![-err_tol.clone(), err_tol])
-                } else {
-                    (res, vec![x])
-                }
-            }
-            HybridOp::ReduceArgMax { dim } => {
-                let res = tensor::ops::argmax_axes(&x, *dim)?;
-                let inter =
-                    Op::f(&HybridOp::ReduceMax { axes: vec![*dim] }, inputs)?.intermediate_lookups;
-
-                (res, inter)
-            }
-            HybridOp::ReduceArgMin { dim } => {
-                let res = tensor::ops::argmin_axes(&x, *dim)?;
-                let inter =
-                    Op::f(&HybridOp::ReduceMin { axes: vec![*dim] }, inputs)?.intermediate_lookups;
-
-                (res, inter)
-            }
+                ..
+            } => crate::tensor::ops::nonlinearities::recip(
+                &x,
+                input_scale.0 as f64,
+                output_scale.0 as f64,
+            ),
+            HybridOp::ReduceArgMax { dim } => tensor::ops::argmax_axes(&x, *dim)?,
+            HybridOp::ReduceArgMin { dim } => tensor::ops::argmin_axes(&x, *dim)?,
             HybridOp::Gather { dim, constant_idx } => {
                 if let Some(idx) = constant_idx {
-                    log::debug!("idx: {}", idx.show());
-                    let res = tensor::ops::gather(&x, idx, *dim)?;
-                    (res.clone(), vec![])
+                    tensor::ops::gather(&x, idx, *dim)?
                 } else {
                     let y = inputs[1].clone().map(|x| felt_to_i128(x));
-                    let res = tensor::ops::gather(&x, &y.map(|x| x as usize), *dim)?;
-                    (res.clone(), vec![])
+                    tensor::ops::gather(&x, &y.map(|x| x as usize), *dim)?
                 }
             }
-            HybridOp::OneHot { dim, num_classes } => (
-                tensor::ops::one_hot(&x, *num_classes, *dim)?.clone(),
-                vec![],
-            ),
-            HybridOp::TopK { dim, k, largest } => {
-                let res = tensor::ops::topk_axes(&x, *k, *dim, *largest)?;
-
-                let mut inter_equals = x
-                    .clone()
-                    .into_iter()
-                    .flat_map(|elem| {
-                        tensor::ops::equals(&res, &vec![elem].into_iter().into())
-                            .unwrap()
-                            .1
-                    })
-                    .collect::<Vec<_>>();
-
-                // sort in descending order and take pairwise differences
-                inter_equals.push(
-                    x.into_iter()
-                        .sorted()
-                        .tuple_windows()
-                        .map(|(a, b)| b - a)
-                        .into(),
-                );
-
-                (res.clone(), inter_equals)
-            }
-            HybridOp::GatherElements { dim, constant_idx } => {
-                if let Some(idx) = constant_idx {
-                    log::debug!("idx: {}", idx.show());
-                    let res = tensor::ops::gather_elements(&x, idx, *dim)?;
-                    (res.clone(), vec![])
-                } else {
-                    let y = inputs[1].clone().map(|x| felt_to_i128(x));
-                    let res = tensor::ops::gather_elements(&x, &y.map(|x| x as usize), *dim)?;
-                    (res.clone(), vec![])
-                }
-            }
-            HybridOp::ScatterElements { dim, constant_idx } => {
-                if let Some(idx) = constant_idx {
-                    log::debug!("idx: {}", idx.show());
-                    let src = inputs[1].clone().map(|x| felt_to_i128(x));
-                    let res = tensor::ops::scatter(&x, idx, &src, *dim)?;
-                    (res.clone(), vec![])
-                } else {
-                    let idx = inputs[1].clone().map(|x| felt_to_i128(x) as usize);
-                    let src = inputs[2].clone().map(|x| felt_to_i128(x));
-                    let res = tensor::ops::scatter(&x, &idx, &src, *dim)?;
-                    (res.clone(), vec![])
-                }
+            HybridOp::OneHot { dim, num_classes } => {
+                tensor::ops::one_hot(&x, *num_classes, *dim)?.clone()
             }
+
+            HybridOp::TopK { dim, k, largest } => tensor::ops::topk_axes(&x, *k, *dim, *largest)?,
             HybridOp::MaxPool2d {
                 padding,
                 stride,
                 pool_dims,
                 ..
-            } => {
-                let max_minus_one =
-                    Tensor::from(vec![x.clone().into_iter().max().unwrap() - 1].into_iter());
-                let unit = Tensor::from(vec![1].into_iter());
-                // relu(x - max(x - 1)
-                let inter_1 = (x.clone() - max_minus_one)?;
-                // relu(1 - sum(relu(inter_1)))
-                let inter_2 = (unit
-                    - tensor::ops::sum(&tensor::ops::nonlinearities::leakyrelu(&inter_1, 0.0))?)?;
-                (
-                    tensor::ops::max_pool2d(&x, padding, stride, pool_dims)?,
-                    vec![inter_1, inter_2],
-                )
-            }
+            } => tensor::ops::max_pool2d(&x, padding, stride, pool_dims)?,
             HybridOp::SumPool {
                 padding,
                 stride,
@@ -265,10 +135,7 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for HybridOp {
             }
             HybridOp::RangeCheck(tol) => {
                 let y = inputs[1].clone().map(|x| felt_to_i128(x));
-                (
-                    tensor::ops::nonlinearities::range_check_percent(&[x, y], 128, 128, tol.val),
-                    vec![],
-                )
+                tensor::ops::nonlinearities::range_check_percent(&[x, y], 128, 128, tol.val)
             }
             HybridOp::Greater => {
                 let y = inputs[1].clone().map(|x| felt_to_i128(x));
@@ -295,10 +162,7 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for HybridOp {
         // convert back to felt
         let output = res.map(|x| i128_to_felt(x));
 
-        Ok(ForwardResult {
-            output,
-            intermediate_lookups,
-        })
+        Ok(ForwardResult { output })
     }
 
     fn as_string(&self) -> String {
@@ -352,8 +216,6 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for HybridOp {
             HybridOp::TopK { k, dim, largest } => {
                 format!("TOPK (k={}, dim={}, largest={})", k, dim, largest)
             }
-            HybridOp::GatherElements { dim, .. } => format!("GATHERELEMENTS (dim={})", dim),
-            HybridOp::ScatterElements { dim, .. } => format!("SCATTERELEMENTS (dim={})", dim),
             HybridOp::OneHot { dim, num_classes } => {
                 format!("ONEHOT (dim={}, num_classes={})", dim, num_classes)
             }
@@ -426,9 +288,7 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for HybridOp {
                         config,
                         region,
                         values.try_into()?,
-                        &LookupOp::Div {
-                            denom: *denom,
-                        },
+                        &LookupOp::Div { denom: *denom },
                     )?
                 }
             }
@@ -439,26 +299,7 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for HybridOp {
                     layouts::gather(config, region, values[..].try_into()?, *dim)?
                 }
             }
-            HybridOp::GatherElements { dim, constant_idx } => {
-                if let Some(idx) = constant_idx {
-                    tensor::ops::gather_elements(values[0].get_inner_tensor()?, idx, *dim)?.into()
-                } else {
-                    layouts::gather_elements(config, region, values[..].try_into()?, *dim)?
-                }
-            }
-            HybridOp::ScatterElements { dim, constant_idx } => {
-                if let Some(idx) = constant_idx {
-                    tensor::ops::scatter(
-                        values[0].get_inner_tensor()?,
-                        idx,
-                        values[1].get_inner_tensor()?,
-                        *dim,
-                    )?
-                    .into()
-                } else {
-                    layouts::scatter_elements(config, region, values[..].try_into()?, *dim)?
-                }
-            }
+
             HybridOp::MaxPool2d {
                 padding,
                 stride,

src/circuit/ops/layouts.rs

@@ -128,13 +128,7 @@ pub fn recip<F: PrimeField + TensorType + PartialOrd>(
 
     let range_check_bracket = felt_to_i128(output_scale * input_scale) / 2;
 
-    let mut scaled_unit =
-        Tensor::from(vec![ValType::Constant(output_scale * input_scale)].into_iter());
-    scaled_unit.set_visibility(&crate::graph::Visibility::Fixed);
-    let scaled_unit = region.assign(&config.inputs[1], &scaled_unit.into())?;
-    region.increment(scaled_unit.len());
-
-    let is_assigned = !input.any_unknowns()? && !scaled_unit.any_unknowns()?;
+    let is_assigned = !input.any_unknowns()?;
 
     let mut claimed_output: ValTensor<F> = if is_assigned {
         let input_evals = input.get_int_evals()?;
@@ -166,27 +160,14 @@ pub fn recip<F: PrimeField + TensorType + PartialOrd>(
 
     log::debug!("product: {:?}", product.get_int_evals()?);
 
-    // this is now of scale 2 * scale hence why we rescaled the unit scale
-    let diff_with_input = pairwise(
-        config,
-        region,
-        &[product.clone(), scaled_unit.clone()],
-        BaseOp::Sub,
-    )?;
-
-    log::debug!("scaled_unit: {:?}", scaled_unit.get_int_evals()?);
-
-    // debug print the diff
-    log::debug!("diff_with_input: {:?}", diff_with_input.get_int_evals()?);
-
     log::debug!("range_check_bracket: {:?}", range_check_bracket);
 
     // at most the error should be in the original unit scale's range
     range_check(
         config,
         region,
-        &[diff_with_input],
-        &(-range_check_bracket, range_check_bracket),
+        &[product],
+        &(range_check_bracket, 3 * range_check_bracket),
     )?;
 
     Ok(claimed_output)
@@ -233,7 +214,7 @@ pub fn dot<F: PrimeField + TensorType + PartialOrd>(
 
     let mut assigned_len = 0;
     for (i, input) in values.iter_mut().enumerate() {
-        input.pad_to_zero_rem(block_width)?;
+        input.pad_to_zero_rem(block_width, ValType::Constant(F::ZERO))?;
         let inp = {
             let (res, len) = region.assign_with_duplication(
                 &config.inputs[i],
@@ -1161,7 +1142,7 @@ pub fn sum<F: PrimeField + TensorType + PartialOrd>(
     let assigned_len: usize;
     let input = {
         let mut input = values[0].clone();
-        input.pad_to_zero_rem(block_width)?;
+        input.pad_to_zero_rem(block_width, ValType::Constant(F::ZERO))?;
         let (res, len) =
             region.assign_with_duplication(&config.inputs[1], &input, &config.check_mode, false)?;
         assigned_len = len;
@@ -1230,7 +1211,7 @@ pub fn prod<F: PrimeField + TensorType + PartialOrd>(
     let assigned_len: usize;
     let input = {
         let mut input = values[0].clone();
-        input.pad_to_zero_rem(block_width)?;
+        input.pad_to_zero_rem(block_width, ValType::Constant(F::ONE))?;
         let (res, len) =
             region.assign_with_duplication(&config.inputs[1], &input, &config.check_mode, false)?;
         assigned_len = len;
@@ -2488,7 +2469,7 @@ pub fn range_check<F: PrimeField + TensorType + PartialOrd>(
     values: &[ValTensor<F>; 1],
     range: &crate::circuit::table::Range,
 ) -> Result<ValTensor<F>, Box<dyn Error>> {
-    region.add_used_range_check(*range);
+    region.add_used_range_check(*range)?;
 
     // time the entire operation
     let timer = instant::Instant::now();
@@ -2534,7 +2515,7 @@ pub fn nonlinearity<F: PrimeField + TensorType + PartialOrd>(
     values: &[ValTensor<F>; 1],
     nl: &LookupOp,
 ) -> Result<ValTensor<F>, Box<dyn Error>> {
-    region.add_used_lookup(nl.clone());
+    region.add_used_lookup(nl.clone(), values)?;
 
     // time the entire operation
     let timer = instant::Instant::now();

src/circuit/ops/lookup.rs

@@ -227,10 +227,7 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for LookupOp {
 
         let output = res.map(|x| i128_to_felt(x));
 
-        Ok(ForwardResult {
-            output,
-            intermediate_lookups: vec![],
-        })
+        Ok(ForwardResult { output })
     }
 
     /// Returns the name of the operation

src/circuit/ops/mod.rs

@@ -29,7 +29,6 @@ pub mod region;
 #[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord, Serialize, Deserialize)]
 pub struct ForwardResult<F: PrimeField + TensorType + PartialOrd> {
     pub(crate) output: Tensor<F>,
-    pub(crate) intermediate_lookups: Vec<Tensor<i128>>,
 }
 
 /// A trait representing operations that can be represented as constraints in a circuit.
@@ -178,7 +177,6 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for Input {
     fn f(&self, x: &[Tensor<F>]) -> Result<ForwardResult<F>, TensorError> {
         Ok(ForwardResult {
             output: x[0].clone(),
-            intermediate_lookups: vec![],
         })
     }
 
@@ -304,10 +302,7 @@ impl<F: PrimeField + TensorType + PartialOrd + Serialize + for<'de> Deserialize<
     fn f(&self, _: &[Tensor<F>]) -> Result<ForwardResult<F>, TensorError> {
         let output = self.quantized_values.clone();
 
-        Ok(ForwardResult {
-            output,
-            intermediate_lookups: vec![],
-        })
+        Ok(ForwardResult { output })
     }
 
     fn as_string(&self) -> String {

src/circuit/ops/poly.rs

@@ -1,5 +1,6 @@
 use crate::{
     circuit::layouts,
+    fieldutils::felt_to_i128,
     tensor::{self, Tensor, TensorError},
 };
 
@@ -9,6 +10,14 @@ use super::{base::BaseOp, *};
 /// An enum representing the operations that can be expressed as arithmetic (non lookup) operations.
 #[derive(Clone, Debug, Serialize, Deserialize)]
 pub enum PolyOp {
+    GatherElements {
+        dim: usize,
+        constant_idx: Option<Tensor<usize>>,
+    },
+    ScatterElements {
+        dim: usize,
+        constant_idx: Option<Tensor<usize>>,
+    },
     MultiBroadcastTo {
         shape: Vec<usize>,
     },
@@ -81,6 +90,8 @@ impl<F: PrimeField + TensorType + PartialOrd + Serialize + for<'de> Deserialize<
 
     fn as_string(&self) -> String {
         match &self {
+            PolyOp::GatherElements { dim, .. } => format!("GATHERELEMENTS (dim={})", dim),
+            PolyOp::ScatterElements { dim, .. } => format!("SCATTERELEMENTS (dim={})", dim),
             PolyOp::MultiBroadcastTo { shape } => format!("MULTIBROADCASTTO (shape={:?})", shape),
             PolyOp::MoveAxis { .. } => "MOVEAXIS".into(),
             PolyOp::Downsample { .. } => "DOWNSAMPLE".into(),
@@ -203,14 +214,36 @@ impl<F: PrimeField + TensorType + PartialOrd + Serialize + for<'de> Deserialize<
                 if 1 != inputs.len() {
                     return Err(TensorError::DimMismatch("slice inputs".to_string()));
                 }
-                Ok(tensor::ops::slice(&inputs[0], axis, start, end)?)
+                tensor::ops::slice(&inputs[0], axis, start, end)
+            }
+            PolyOp::GatherElements { dim, constant_idx } => {
+                let x = inputs[0].clone();
+                let y = if let Some(idx) = constant_idx {
+                    idx.clone()
+                } else {
+                    inputs[1].clone().map(|x| felt_to_i128(x) as usize)
+                };
+                tensor::ops::gather_elements(&x, &y, *dim)
+            }
+            PolyOp::ScatterElements { dim, constant_idx } => {
+                let x = inputs[0].clone();
+
+                let idx = if let Some(idx) = constant_idx {
+                    idx.clone()
+                } else {
+                    inputs[1].clone().map(|x| felt_to_i128(x) as usize)
+                };
+
+                let src = if let Some(_) = constant_idx {
+                    inputs[1].clone()
+                } else {
+                    inputs[2].clone()
+                };
+                tensor::ops::scatter(&x, &idx, &src, *dim)
             }
         }?;
 
-        Ok(ForwardResult {
-            output: res,
-            intermediate_lookups: vec![],
-        })
+        Ok(ForwardResult { output: res })
     }
 
     fn layout(
@@ -251,6 +284,26 @@ impl<F: PrimeField + TensorType + PartialOrd + Serialize + for<'de> Deserialize<
             PolyOp::Conv { padding, stride } => {
                 layouts::conv(config, region, values[..].try_into()?, *padding, *stride)?
             }
+            PolyOp::GatherElements { dim, constant_idx } => {
+                if let Some(idx) = constant_idx {
+                    tensor::ops::gather_elements(values[0].get_inner_tensor()?, idx, *dim)?.into()
+                } else {
+                    layouts::gather_elements(config, region, values[..].try_into()?, *dim)?
+                }
+            }
+            PolyOp::ScatterElements { dim, constant_idx } => {
+                if let Some(idx) = constant_idx {
+                    tensor::ops::scatter(
+                        values[0].get_inner_tensor()?,
+                        idx,
+                        values[1].get_inner_tensor()?,
+                        *dim,
+                    )?
+                    .into()
+                } else {
+                    layouts::scatter_elements(config, region, values[..].try_into()?, *dim)?
+                }
+            }
             PolyOp::DeConv {
                 padding,
                 output_padding,
@@ -352,6 +405,8 @@ impl<F: PrimeField + TensorType + PartialOrd + Serialize + for<'de> Deserialize<
             vec![1, 2]
         } else if matches!(self, PolyOp::Concat { .. }) {
             (0..100).collect()
+        } else if matches!(self, PolyOp::ScatterElements { .. }) {
+            vec![0, 2]
         } else {
             vec![]
         }

src/circuit/ops/region.rs

@@ -16,6 +16,8 @@ use std::{
     },
 };
 
+use portable_atomic::AtomicI128 as AtomicInt;
+
 use super::lookup::LookupOp;
 
 /// Region error
@@ -66,6 +68,8 @@ pub struct RegionCtx<'a, F: PrimeField + TensorType + PartialOrd> {
     total_constants: usize,
     used_lookups: HashSet<LookupOp>,
     used_range_checks: HashSet<Range>,
+    max_lookup_inputs: i128,
+    min_lookup_inputs: i128,
 }
 
 impl<'a, F: PrimeField + TensorType + PartialOrd> RegionCtx<'a, F> {
@@ -87,6 +91,8 @@ impl<'a, F: PrimeField + TensorType + PartialOrd> RegionCtx<'a, F> {
             total_constants: 0,
             used_lookups: HashSet::new(),
             used_range_checks: HashSet::new(),
+            max_lookup_inputs: 0,
+            min_lookup_inputs: 0,
         }
     }
     /// Create a new region context from a wrapped region
@@ -104,6 +110,8 @@ impl<'a, F: PrimeField + TensorType + PartialOrd> RegionCtx<'a, F> {
             total_constants: 0,
             used_lookups: HashSet::new(),
             used_range_checks: HashSet::new(),
+            max_lookup_inputs: 0,
+            min_lookup_inputs: 0,
         }
     }
 
@@ -120,6 +128,8 @@ impl<'a, F: PrimeField + TensorType + PartialOrd> RegionCtx<'a, F> {
             total_constants: 0,
             used_lookups: HashSet::new(),
             used_range_checks: HashSet::new(),
+            max_lookup_inputs: 0,
+            min_lookup_inputs: 0,
         }
     }
 
@@ -141,6 +151,8 @@ impl<'a, F: PrimeField + TensorType + PartialOrd> RegionCtx<'a, F> {
             total_constants,
             used_lookups,
             used_range_checks,
+            max_lookup_inputs: 0,
+            min_lookup_inputs: 0,
         }
     }
 
@@ -180,6 +192,7 @@ impl<'a, F: PrimeField + TensorType + PartialOrd> RegionCtx<'a, F> {
 
     /// Create a new region context per loop iteration
     /// hacky but it works
+
     pub fn dummy_loop<T: TensorType + Send + Sync>(
         &mut self,
         output: &mut Tensor<T>,
@@ -190,6 +203,10 @@ impl<'a, F: PrimeField + TensorType + PartialOrd> RegionCtx<'a, F> {
         let row = AtomicUsize::new(self.row());
         let linear_coord = AtomicUsize::new(self.linear_coord());
         let constants = AtomicUsize::new(self.total_constants());
+        let max_lookup_inputs =
+            AtomicInt::new(self.max_lookup_inputs().try_into().unwrap_or_default());
+        let min_lookup_inputs =
+            AtomicInt::new(self.min_lookup_inputs().try_into().unwrap_or_default());
         let lookups = Arc::new(Mutex::new(self.used_lookups.clone()));
         let range_checks = Arc::new(Mutex::new(self.used_range_checks.clone()));
 
@@ -221,6 +238,14 @@ impl<'a, F: PrimeField + TensorType + PartialOrd> RegionCtx<'a, F> {
                     local_reg.total_constants() - starting_constants,
                     Ordering::SeqCst,
                 );
+
+                let local_max_lookup_inputs =
+                    local_reg.max_lookup_inputs().try_into().unwrap_or_default();
+                let local_min_lookup_inputs =
+                    local_reg.min_lookup_inputs().try_into().unwrap_or_default();
+
+                max_lookup_inputs.fetch_max(local_max_lookup_inputs, Ordering::SeqCst);
+                min_lookup_inputs.fetch_min(local_min_lookup_inputs, Ordering::SeqCst);
                 // update the lookups
                 let mut lookups = lookups.lock().unwrap();
                 lookups.extend(local_reg.used_lookups());
@@ -234,6 +259,11 @@ impl<'a, F: PrimeField + TensorType + PartialOrd> RegionCtx<'a, F> {
             })?;
         self.total_constants = constants.into_inner();
         self.linear_coord = linear_coord.into_inner();
+        #[allow(trivial_numeric_casts)]
+        {
+            self.max_lookup_inputs = max_lookup_inputs.into_inner() as i128;
+            self.min_lookup_inputs = min_lookup_inputs.into_inner() as i128;
+        }
         self.row = row.into_inner();
         self.used_lookups = Arc::try_unwrap(lookups)
             .map_err(|e| RegionError::from(format!("dummy_loop: failed to get lookups: {:?}", e)))?
@@ -253,19 +283,54 @@ impl<'a, F: PrimeField + TensorType + PartialOrd> RegionCtx<'a, F> {
         Ok(())
     }
 
+    /// Update the max and min from inputs
+    pub fn update_max_min_lookup_inputs(
+        &mut self,
+        inputs: &[ValTensor<F>],
+    ) -> Result<(), Box<dyn std::error::Error>> {
+        let (mut min, mut max) = (0, 0);
+        for i in inputs {
+            max = max.max(i.get_int_evals()?.into_iter().max().unwrap_or_default());
+            min = min.min(i.get_int_evals()?.into_iter().min().unwrap_or_default());
+        }
+        self.max_lookup_inputs = self.max_lookup_inputs.max(max);
+        self.min_lookup_inputs = self.min_lookup_inputs.min(min);
+        Ok(())
+    }
+
+    /// Update the max and min from inputs
+    pub fn update_max_min_lookup_range(
+        &mut self,
+        range: Range,
+    ) -> Result<(), Box<dyn std::error::Error>> {
+        if range.0 > range.1 {
+            return Err("update_max_min_lookup_range: invalid range".into());
+        }
+
+        self.max_lookup_inputs = self.max_lookup_inputs.max(range.1);
+        self.min_lookup_inputs = self.min_lookup_inputs.min(range.0);
+        Ok(())
+    }
+
     /// Check if the region is dummy
     pub fn is_dummy(&self) -> bool {
         self.region.is_none()
     }
 
     /// add used lookup
-    pub fn add_used_lookup(&mut self, lookup: LookupOp) {
+    pub fn add_used_lookup(
+        &mut self,
+        lookup: LookupOp,
+        inputs: &[ValTensor<F>],
+    ) -> Result<(), Box<dyn std::error::Error>> {
         self.used_lookups.insert(lookup);
+        self.update_max_min_lookup_inputs(inputs)
     }
 
     /// add used range check
-    pub fn add_used_range_check(&mut self, range: Range) {
+    pub fn add_used_range_check(&mut self, range: Range) -> Result<(), Box<dyn std::error::Error>> {
         self.used_range_checks.insert(range);
+        self.update_max_min_lookup_range(range)
     }
 
     /// Get the offset
@@ -293,6 +358,16 @@ impl<'a, F: PrimeField + TensorType + PartialOrd> RegionCtx<'a, F> {
         self.used_range_checks.clone()
     }
 
+    /// max lookup inputs
+    pub fn max_lookup_inputs(&self) -> i128 {
+        self.max_lookup_inputs
+    }
+
+    /// min lookup inputs
+    pub fn min_lookup_inputs(&self) -> i128 {
+        self.min_lookup_inputs
+    }
+
     /// Assign a constant value
     pub fn assign_constant(&mut self, var: &VarTensor, value: F) -> Result<ValType<F>, Error> {
         self.total_constants += 1;

src/circuit/tests.rs

@@ -246,7 +246,7 @@ mod matmul_col_overflow {
 #[cfg(test)]
 #[cfg(not(all(target_arch = "wasm32", target_os = "unknown")))]
 mod matmul_col_ultra_overflow_double_col {
-    use halo2_proofs::poly::commitment::ParamsProver;
+    use halo2_proofs::poly::commitment::{Params, ParamsProver};
 
     use super::*;
 
@@ -349,8 +349,13 @@ mod matmul_col_ultra_overflow_double_col {
         let strategy =
             halo2_proofs::poly::kzg::strategy::SingleStrategy::new(params.verifier_params());
         let vk = pk.get_vk();
-        let result =
-            crate::pfsys::verify_proof_circuit_kzg(params.verifier_params(), proof, vk, strategy);
+        let result = crate::pfsys::verify_proof_circuit_kzg(
+            params.verifier_params(),
+            proof,
+            vk,
+            strategy,
+            params.n(),
+        );
 
         assert!(result.is_ok());
 
@@ -361,7 +366,7 @@ mod matmul_col_ultra_overflow_double_col {
 #[cfg(test)]
 #[cfg(not(all(target_arch = "wasm32", target_os = "unknown")))]
 mod matmul_col_ultra_overflow {
-    use halo2_proofs::poly::commitment::ParamsProver;
+    use halo2_proofs::poly::commitment::{Params, ParamsProver};
 
     use super::*;
 
@@ -463,8 +468,13 @@ mod matmul_col_ultra_overflow {
         let strategy =
             halo2_proofs::poly::kzg::strategy::SingleStrategy::new(params.verifier_params());
         let vk = pk.get_vk();
-        let result =
-            crate::pfsys::verify_proof_circuit_kzg(params.verifier_params(), proof, vk, strategy);
+        let result = crate::pfsys::verify_proof_circuit_kzg(
+            params.verifier_params(),
+            proof,
+            vk,
+            strategy,
+            params.n(),
+        );
 
         assert!(result.is_ok());
 
@@ -1140,7 +1150,7 @@ mod conv {
 #[cfg(test)]
 #[cfg(not(all(target_arch = "wasm32", target_os = "unknown")))]
 mod conv_col_ultra_overflow {
-    use halo2_proofs::poly::commitment::ParamsProver;
+    use halo2_proofs::poly::commitment::{Params, ParamsProver};
 
     use super::*;
 
@@ -1262,8 +1272,13 @@ mod conv_col_ultra_overflow {
         let strategy =
             halo2_proofs::poly::kzg::strategy::SingleStrategy::new(params.verifier_params());
         let vk = pk.get_vk();
-        let result =
-            crate::pfsys::verify_proof_circuit_kzg(params.verifier_params(), proof, vk, strategy);
+        let result = crate::pfsys::verify_proof_circuit_kzg(
+            params.verifier_params(),
+            proof,
+            vk,
+            strategy,
+            params.n(),
+        );
 
         assert!(result.is_ok());
 
@@ -1275,7 +1290,7 @@ mod conv_col_ultra_overflow {
 // not wasm 32 unknown
 #[cfg(not(all(target_arch = "wasm32", target_os = "unknown")))]
 mod conv_relu_col_ultra_overflow {
-    use halo2_proofs::poly::commitment::ParamsProver;
+    use halo2_proofs::poly::commitment::{Params, ParamsProver};
 
     use super::*;
 
@@ -1412,8 +1427,13 @@ mod conv_relu_col_ultra_overflow {
         let strategy =
             halo2_proofs::poly::kzg::strategy::SingleStrategy::new(params.verifier_params());
         let vk = pk.get_vk();
-        let result =
-            crate::pfsys::verify_proof_circuit_kzg(params.verifier_params(), proof, vk, strategy);
+        let result = crate::pfsys::verify_proof_circuit_kzg(
+            params.verifier_params(),
+            proof,
+            vk,
+            strategy,
+            params.n(),
+        );
 
         assert!(result.is_ok());
 
@@ -2343,7 +2363,7 @@ mod lookup_ultra_overflow {
     use halo2_proofs::{
         circuit::{Layouter, SimpleFloorPlanner, Value},
         plonk::{Circuit, ConstraintSystem, Error},
-        poly::commitment::ParamsProver,
+        poly::commitment::{Params, ParamsProver},
     };
 
     #[derive(Clone)]
@@ -2447,8 +2467,13 @@ mod lookup_ultra_overflow {
         let strategy =
             halo2_proofs::poly::kzg::strategy::SingleStrategy::new(params.verifier_params());
         let vk = pk.get_vk();
-        let result =
-            crate::pfsys::verify_proof_circuit_kzg(params.verifier_params(), proof, vk, strategy);
+        let result = crate::pfsys::verify_proof_circuit_kzg(
+            params.verifier_params(),
+            proof,
+            vk,
+            strategy,
+            params.n(),
+        );
 
         assert!(result.is_ok());
 

src/commands.rs

@@ -85,6 +85,8 @@ pub const DEFAULT_VK_SOL: &str = "vk.sol";
 pub const DEFAULT_VK_ABI: &str = "vk.abi";
 /// Default scale rebase multipliers for calibration
 pub const DEFAULT_SCALE_REBASE_MULTIPLIERS: &str = "1,2,10";
+/// Default use reduced srs for verification
+pub const DEFAULT_USE_REDUCED_SRS_FOR_VERIFICATION: &str = "false";
 
 impl std::fmt::Display for TranscriptType {
     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
@@ -201,7 +203,7 @@ impl<'source> FromPyObject<'source> for CalibrationTarget {
         }
     }
 }
-
+// not wasm
 use lazy_static::lazy_static;
 
 // if CARGO VERSION is 0.0.0 replace with "source - no compatibility guaranteed"
@@ -695,6 +697,9 @@ pub enum Commands {
         /// The path to SRS, if None will use $EZKL_REPO_PATH/srs/kzg{logrows}.srs
         #[arg(long)]
         srs_path: Option<PathBuf>,
+        /// Reduce SRS logrows to the number of instances rather than the number of logrows used for proofs (only works if the srs were generated in the same ceremony)
+        #[arg(long, default_value = DEFAULT_USE_REDUCED_SRS_FOR_VERIFICATION)]
+        reduced_srs: Option<bool>,
     },
     /// Verifies an aggregate proof, returning accept or reject
     VerifyAggr {
@@ -795,12 +800,4 @@ pub enum Commands {
         #[arg(long)]
         addr_vk: Option<H160>,
     },
-
-    /// Print the proof in hexadecimal
-    #[command(name = "print-proof-hex")]
-    PrintProofHex {
-        /// The path to the proof file
-        #[arg(long, default_value = DEFAULT_PROOF)]
-        proof_path: PathBuf,
-    },
 }

src/execute.rs

@@ -366,7 +366,8 @@ pub async fn run(command: Commands) -> Result<String, Box<dyn Error>> {
             settings_path,
             vk_path,
             srs_path,
-        } => verify(proof_path, settings_path, vk_path, srs_path)
+            reduced_srs,
+        } => verify(proof_path, settings_path, vk_path, srs_path, reduced_srs)
             .map(|e| serde_json::to_string(&e).unwrap()),
         Commands::VerifyAggr {
             proof_path,
@@ -440,7 +441,6 @@ pub async fn run(command: Commands) -> Result<String, Box<dyn Error>> {
             addr_da,
             addr_vk,
         } => verify_evm(proof_path, addr_verifier, rpc_url, addr_da, addr_vk).await,
-        Commands::PrintProofHex { proof_path } => print_proof_hex(proof_path),
     }
 }
 
@@ -628,7 +628,7 @@ pub(crate) async fn gen_witness(
     );
 
     if let Some(output_path) = output {
-        serde_json::to_writer(&File::create(output_path)?, &witness)?;
+        witness.save(output_path)?;
     }
 
     // print the witness in debug
@@ -826,10 +826,7 @@ pub(crate) fn calibrate(
     let range = if let Some(scales) = scales {
         scales
     } else {
-        match target {
-            CalibrationTarget::Resources { .. } => (8..10).collect::<Vec<crate::Scale>>(),
-            CalibrationTarget::Accuracy => (10..14).collect::<Vec<crate::Scale>>(),
-        }
+        (10..14).collect::<Vec<crate::Scale>>()
     };
 
     let div_rebasing = if let Some(div_rebasing) = div_rebasing {
@@ -1172,16 +1169,6 @@ pub(crate) fn mock(
     Ok(String::new())
 }
 
-pub(crate) fn print_proof_hex(proof_path: PathBuf) -> Result<String, Box<dyn Error>> {
-    let proof = Snark::load::<KZGCommitmentScheme<Bn256>>(&proof_path)?;
-    for instance in proof.instances {
-        println!("{:?}", instance);
-    }
-    let hex_str = hex::encode(proof.proof);
-    info!("0x{}", hex_str);
-    Ok(format!("0x{}", hex_str))
-}
-
 #[cfg(feature = "render")]
 pub(crate) fn render(
     model: PathBuf,
@@ -1728,6 +1715,7 @@ pub(crate) fn fuzz(
             bad_proof,
             pk.get_vk(),
             strategy.clone(),
+            params.n(),
         )
         .map_err(|_| ())
     };
@@ -1758,6 +1746,7 @@ pub(crate) fn fuzz(
             bad_proof,
             pk.get_vk(),
             strategy.clone(),
+            params.n(),
         )
         .map_err(|_| ())
     };
@@ -1794,6 +1783,7 @@ pub(crate) fn fuzz(
             proof.clone(),
             bad_vk,
             strategy.clone(),
+            params.n(),
         )
         .map_err(|_| ())
     };
@@ -1825,6 +1815,7 @@ pub(crate) fn fuzz(
             bad_proof,
             pk.get_vk(),
             strategy.clone(),
+            params.n(),
         )
         .map_err(|_| ())
     };
@@ -1860,6 +1851,7 @@ pub(crate) fn fuzz(
             bad_proof,
             pk.get_vk(),
             strategy.clone(),
+            params.n(),
         )
         .map_err(|_| ())
     };
@@ -2045,15 +2037,33 @@ pub(crate) fn verify(
     settings_path: PathBuf,
     vk_path: PathBuf,
     srs_path: Option<PathBuf>,
+    reduced_srs: Option<bool>,
 ) -> Result<bool, Box<dyn Error>> {
     let circuit_settings = GraphSettings::load(&settings_path)?;
-    let params = load_params_cmd(srs_path, circuit_settings.run_args.logrows)?;
+
+    let params = if let Some(reduced_srs) = reduced_srs {
+        if reduced_srs {
+            load_params_cmd(srs_path, circuit_settings.log2_total_instances())?
+        } else {
+            load_params_cmd(srs_path, circuit_settings.run_args.logrows)?
+        }
+    } else {
+        load_params_cmd(srs_path, circuit_settings.run_args.logrows)?
+    };
+
     let proof = Snark::load::<KZGCommitmentScheme<Bn256>>(&proof_path)?;
 
     let strategy = KZGSingleStrategy::new(params.verifier_params());
-    let vk = load_vk::<KZGCommitmentScheme<Bn256>, Fr, GraphCircuit>(vk_path, circuit_settings)?;
+    let vk =
+        load_vk::<KZGCommitmentScheme<Bn256>, Fr, GraphCircuit>(vk_path, circuit_settings.clone())?;
     let now = Instant::now();
-    let result = verify_proof_circuit_kzg(params.verifier_params(), proof, &vk, strategy);
+    let result = verify_proof_circuit_kzg(
+        params.verifier_params(),
+        proof,
+        &vk,
+        strategy,
+        1 << circuit_settings.run_args.logrows,
+    );
     let elapsed = now.elapsed();
     info!(
         "verify took {}.{}",
@@ -2077,7 +2087,7 @@ pub(crate) fn verify_aggr(
     let strategy = AccumulatorStrategy::new(params.verifier_params());
     let vk = load_vk::<KZGCommitmentScheme<Bn256>, Fr, AggregationCircuit>(vk_path, ())?;
     let now = Instant::now();
-    let result = verify_proof_circuit_kzg(&params, proof, &vk, strategy);
+    let result = verify_proof_circuit_kzg(&params, proof, &vk, strategy, 1 << logrows);
 
     let elapsed = now.elapsed();
     info!(

src/graph/mod.rs

@@ -28,14 +28,17 @@ use crate::circuit::{CheckMode, InputType};
 use crate::fieldutils::felt_to_f64;
 use crate::pfsys::PrettyElements;
 use crate::tensor::{Tensor, ValTensor};
-use crate::RunArgs;
+use crate::{RunArgs, EZKL_BUF_CAPACITY};
+
 use halo2_proofs::{
     circuit::Layouter,
     plonk::{Circuit, ConstraintSystem, Error as PlonkError},
 };
 use halo2curves::bn256::{self, Bn256, Fr as Fp, G1Affine};
 use halo2curves::ff::PrimeField;
-use log::{debug, error, info, trace, warn};
+#[cfg(not(target_arch = "wasm32"))]
+use lazy_static::lazy_static;
+use log::{debug, error, trace, warn};
 use maybe_rayon::prelude::{IntoParallelRefIterator, ParallelIterator};
 pub use model::*;
 pub use node::*;
@@ -46,7 +49,6 @@ use pyo3::types::PyDict;
 #[cfg(feature = "python-bindings")]
 use pyo3::ToPyObject;
 use serde::{Deserialize, Serialize};
-use std::io::{Read, Write};
 use std::ops::Deref;
 use thiserror::Error;
 pub use utilities::*;
@@ -61,6 +63,20 @@ pub const RANGE_MULTIPLIER: i128 = 2;
 /// Max representation of a lookup table input
 pub const MAX_LOOKUP_ABS: i128 = 8 * 2_i128.pow(MAX_PUBLIC_SRS);
 
+#[cfg(not(target_arch = "wasm32"))]
+lazy_static! {
+    /// Max circuit area
+    pub static ref EZKL_MAX_CIRCUIT_AREA: Option<usize> =
+        if let Ok(max_circuit_area) = std::env::var("EZKL_MAX_CIRCUIT_AREA") {
+            Some(max_circuit_area.parse().unwrap_or(0))
+        } else {
+            None
+        };
+}
+
+#[cfg(target_arch = "wasm32")]
+const EZKL_MAX_CIRCUIT_AREA: Option<usize> = None;
+
 /// circuit related errors.
 #[derive(Debug, Error)]
 pub enum GraphError {
@@ -292,16 +308,20 @@ impl GraphWitness {
 
     /// Load the model input from a file
     pub fn from_path(path: std::path::PathBuf) -> Result<Self, Box<dyn std::error::Error>> {
-        let mut file = std::fs::File::open(path.clone())
+        let file = std::fs::File::open(path.clone())
             .map_err(|_| format!("failed to load model at {}", path.display()))?;
-        let mut data = String::new();
-        file.read_to_string(&mut data)?;
-        serde_json::from_str(&data).map_err(|e| e.into())
+
+        let reader = std::io::BufReader::with_capacity(*EZKL_BUF_CAPACITY, file);
+        serde_json::from_reader(reader).map_err(|e| e.into())
     }
 
     /// Save the model input to a file
     pub fn save(&self, path: std::path::PathBuf) -> Result<(), Box<dyn std::error::Error>> {
-        serde_json::to_writer(std::fs::File::create(path)?, &self).map_err(|e| e.into())
+        // use buf writer
+        let writer =
+            std::io::BufWriter::with_capacity(*EZKL_BUF_CAPACITY, std::fs::File::create(path)?);
+
+        serde_json::to_writer(writer, &self).map_err(|e| e.into())
     }
 
     ///
@@ -456,22 +476,33 @@ impl GraphSettings {
         instances
     }
 
+    /// calculate the log2 of the total number of instances
+    pub fn log2_total_instances(&self) -> u32 {
+        let sum = self.total_instances().iter().sum::<usize>();
+
+        // max between 1 and the log2 of the sums
+        std::cmp::max((sum as f64).log2().ceil() as u32, 1)
+    }
+
     /// save params to file
     pub fn save(&self, path: &std::path::PathBuf) -> Result<(), std::io::Error> {
-        let encoded = serde_json::to_string(&self)?;
-        let mut file = std::fs::File::create(path)?;
-        file.write_all(encoded.as_bytes())
+        // buf writer
+        let writer =
+            std::io::BufWriter::with_capacity(*EZKL_BUF_CAPACITY, std::fs::File::create(path)?);
+        serde_json::to_writer(writer, &self).map_err(|e| {
+            error!("failed to save settings file at {}", e);
+            std::io::Error::new(std::io::ErrorKind::Other, e)
+        })
     }
     /// load params from file
     pub fn load(path: &std::path::PathBuf) -> Result<Self, std::io::Error> {
-        let mut file = std::fs::File::open(path).map_err(|e| {
-            error!("failed to open settings file at {}", e);
-            e
-        })?;
-        let mut data = String::new();
-        file.read_to_string(&mut data)?;
-        let res = serde_json::from_str(&data)?;
-        Ok(res)
+        // buf reader
+        let reader =
+            std::io::BufReader::with_capacity(*EZKL_BUF_CAPACITY, std::fs::File::open(path)?);
+        serde_json::from_reader(reader).map_err(|e| {
+            error!("failed to load settings file at {}", e);
+            std::io::Error::new(std::io::ErrorKind::Other, e)
+        })
     }
 
     /// Export the ezkl configuration as json
@@ -530,6 +561,7 @@ impl GraphSettings {
 pub struct GraphConfig {
     model_config: ModelConfig,
     module_configs: ModuleConfigs,
+    circuit_size: CircuitSize,
 }
 
 /// Defines the circuit for a computational graph / model loaded from a `.onnx` file.
@@ -566,7 +598,7 @@ impl GraphCircuit {
     ///
     pub fn save(&self, path: std::path::PathBuf) -> Result<(), Box<dyn std::error::Error>> {
         let f = std::fs::File::create(path)?;
-        let writer = std::io::BufWriter::new(f);
+        let writer = std::io::BufWriter::with_capacity(*EZKL_BUF_CAPACITY, f);
         bincode::serialize_into(writer, &self)?;
         Ok(())
     }
@@ -574,11 +606,10 @@ impl GraphCircuit {
     ///
     pub fn load(path: std::path::PathBuf) -> Result<Self, Box<dyn std::error::Error>> {
         // read bytes from file
-        let mut f = std::fs::File::open(&path)?;
-        let metadata = std::fs::metadata(&path)?;
-        let mut buffer = vec![0; metadata.len() as usize];
-        f.read_exact(&mut buffer)?;
-        let result = bincode::deserialize(&buffer)?;
+        let f = std::fs::File::open(&path)?;
+        let reader = std::io::BufReader::with_capacity(*EZKL_BUF_CAPACITY, f);
+        let result: GraphCircuit = bincode::deserialize_from(reader)?;
+
         Ok(result)
     }
 }
@@ -809,7 +840,7 @@ impl GraphCircuit {
         let shapes = self.model().graph.input_shapes()?;
         let scales = self.model().graph.get_input_scales();
         let input_types = self.model().graph.get_input_types()?;
-        info!("input scales: {:?}", scales);
+        debug!("input scales: {:?}", scales);
 
         match &data.input_data {
             DataSource::File(file_data) => {
@@ -828,7 +859,7 @@ impl GraphCircuit {
         let shapes = self.model().graph.input_shapes()?;
         let scales = self.model().graph.get_input_scales();
         let input_types = self.model().graph.get_input_types()?;
-        info!("input scales: {:?}", scales);
+        debug!("input scales: {:?}", scales);
 
         self.process_data_source(&data.input_data, shapes, scales, input_types)
             .await
@@ -1028,7 +1059,7 @@ impl GraphCircuit {
                 "extended k is too large to accommodate the quotient polynomial with logrows {}",
                 min_logrows
             );
-            error!("{}", err_string);
+            debug!("{}", err_string);
             return Err(err_string.into());
         }
 
@@ -1048,7 +1079,7 @@ impl GraphCircuit {
                 "extended k is too large to accommodate the quotient polynomial with logrows {}",
                 max_logrows
             );
-            error!("{}", err_string);
+            debug!("{}", err_string);
             return Err(err_string.into());
         }
 
@@ -1114,7 +1145,7 @@ impl GraphCircuit {
 
         settings_mut.run_args.logrows = std::cmp::min(max_logrows, settings_mut.run_args.logrows);
 
-        info!(
+        debug!(
             "setting lookup_range to: {:?}, setting logrows to: {}",
             self.settings().run_args.lookup_range,
             self.settings().run_args.logrows
@@ -1366,7 +1397,6 @@ impl GraphCircuit {
     }
 }
 
-#[cfg(not(target_arch = "wasm32"))]
 #[derive(Clone, Debug, Default, Serialize, Deserialize)]
 struct CircuitSize {
     num_instances: usize,
@@ -1374,20 +1404,22 @@ struct CircuitSize {
     num_fixed: usize,
     num_challenges: usize,
     num_selectors: usize,
+    logrows: u32,
 }
 
-#[cfg(not(target_arch = "wasm32"))]
 impl CircuitSize {
-    pub fn from_cs(cs: &ConstraintSystem<Fp>) -> Self {
+    pub fn from_cs(cs: &ConstraintSystem<Fp>, logrows: u32) -> Self {
         CircuitSize {
             num_instances: cs.num_instance_columns(),
             num_advice_columns: cs.num_advice_columns(),
             num_fixed: cs.num_fixed_columns(),
             num_challenges: cs.num_challenges(),
             num_selectors: cs.num_selectors(),
+            logrows,
         }
     }
 
+    #[cfg(not(target_arch = "wasm32"))]
     /// Export the ezkl configuration as json
     pub fn as_json(&self) -> Result<String, Box<dyn std::error::Error>> {
         let serialized = match serde_json::to_string(&self) {
@@ -1398,6 +1430,25 @@ impl CircuitSize {
         };
         Ok(serialized)
     }
+
+    /// number of columns
+    pub fn num_columns(&self) -> usize {
+        self.num_instances + self.num_advice_columns + self.num_fixed
+    }
+
+    /// area of the circuit
+    pub fn area(&self) -> usize {
+        self.num_columns() * (1 << self.logrows)
+    }
+
+    /// area less than max
+    pub fn area_less_than_max(&self) -> bool {
+        if EZKL_MAX_CIRCUIT_AREA.is_some() {
+            self.area() < EZKL_MAX_CIRCUIT_AREA.unwrap()
+        } else {
+            true
+        }
+    }
 }
 
 impl Circuit<Fp> for GraphCircuit {
@@ -1472,10 +1523,12 @@ impl Circuit<Fp> for GraphCircuit {
             (cs.degree() as f32).log2().ceil()
         );
 
+        let circuit_size = CircuitSize::from_cs(cs, params.run_args.logrows);
+
         #[cfg(not(target_arch = "wasm32"))]
-        info!(
+        debug!(
             "circuit size: \n {}",
-            CircuitSize::from_cs(cs)
+            circuit_size
                 .as_json()
                 .unwrap()
                 .to_colored_json_auto()
@@ -1485,6 +1538,7 @@ impl Circuit<Fp> for GraphCircuit {
         GraphConfig {
             model_config,
             module_configs,
+            circuit_size,
         }
     }
 
@@ -1497,6 +1551,16 @@ impl Circuit<Fp> for GraphCircuit {
         config: Self::Config,
         mut layouter: impl Layouter<Fp>,
     ) -> Result<(), PlonkError> {
+        // check if the circuit area is less than the max
+        if !config.circuit_size.area_less_than_max() {
+            error!(
+                "circuit area {} is larger than the max allowed area {}",
+                config.circuit_size.area(),
+                EZKL_MAX_CIRCUIT_AREA.unwrap()
+            );
+            return Err(PlonkError::Synthesis);
+        }
+
         trace!("Setting input in synthesize");
         let input_vis = &self.settings().run_args.input_visibility;
         let output_vis = &self.settings().run_args.output_visibility;

src/graph/model.rs

@@ -10,7 +10,6 @@ use crate::circuit::table::Range;
 use crate::circuit::Input;
 use crate::circuit::InputType;
 use crate::circuit::Unknown;
-use crate::fieldutils::felt_to_i128;
 use crate::tensor::ValType;
 use crate::{
     circuit::{lookup::LookupOp, BaseConfig as PolyConfig, CheckMode, Op},
@@ -68,6 +67,16 @@ pub struct ForwardResult {
     pub min_lookup_inputs: i128,
 }
 
+impl From<DummyPassRes> for ForwardResult {
+    fn from(res: DummyPassRes) -> Self {
+        Self {
+            outputs: res.outputs,
+            max_lookup_inputs: res.max_lookup_inputs,
+            min_lookup_inputs: res.min_lookup_inputs,
+        }
+    }
+}
+
 /// A circuit configuration for the entirety of a model loaded from an Onnx file.
 #[derive(Clone, Debug)]
 pub struct ModelConfig {
@@ -93,6 +102,12 @@ pub struct DummyPassRes {
     pub lookup_ops: HashSet<LookupOp>,
     /// range checks
     pub range_checks: HashSet<Range>,
+    /// max lookup inputs
+    pub max_lookup_inputs: i128,
+    /// min lookup inputs
+    pub min_lookup_inputs: i128,
+    /// outputs
+    pub outputs: Vec<Tensor<Fp>>,
 }
 
 /// A struct for loading from an Onnx file and converting a computational graph to a circuit.
@@ -478,7 +493,7 @@ impl Model {
     ) -> Result<GraphSettings, Box<dyn Error>> {
         let instance_shapes = self.instance_shapes()?;
         #[cfg(not(target_arch = "wasm32"))]
-        info!(
+        debug!(
             "{} {} {}",
             "model has".blue(),
             instance_shapes.len().to_string().blue(),
@@ -486,7 +501,25 @@ impl Model {
         );
         // this is the total number of variables we will need to allocate
         // for the circuit
-        let res = self.dummy_layout(run_args, &self.graph.input_shapes()?)?;
+        let default_value = if !self.visibility.input.is_fixed() {
+            ValType::Value(Value::<Fp>::unknown())
+        } else {
+            ValType::Constant(Fp::ONE)
+        };
+
+        let inputs: Vec<ValTensor<Fp>> = self
+            .graph
+            .input_shapes()?
+            .iter()
+            .map(|shape| {
+                let mut t: ValTensor<Fp> =
+                    vec![default_value.clone(); shape.iter().product()].into();
+                t.reshape(shape)?;
+                Ok(t)
+            })
+            .collect::<Result<Vec<_>, Box<dyn Error>>>()?;
+
+        let res = self.dummy_layout(run_args, &inputs)?;
 
         // if we're using percentage tolerance, we need to add the necessary range check ops for it.
 
@@ -522,206 +555,12 @@ impl Model {
     /// * `model_inputs` - A vector of [Tensor]s to use as inputs to the model.
     /// * `run_args` - [RunArgs]
     pub fn forward(&self, model_inputs: &[Tensor<Fp>]) -> Result<ForwardResult, Box<dyn Error>> {
-        let mut results: BTreeMap<&usize, Vec<Tensor<Fp>>> = BTreeMap::new();
-        let mut max_lookup_inputs = 0;
-        let mut min_lookup_inputs = 0;
-
-        let input_shapes = self.graph.input_shapes()?;
-
-        for (i, input_idx) in self.graph.inputs.iter().enumerate() {
-            let mut input = model_inputs[i].clone();
-            input.reshape(&input_shapes[i])?;
-            results.insert(input_idx, vec![input]);
-        }
-
-        for (idx, n) in self.graph.nodes.iter() {
-            let mut inputs = vec![];
-            if n.is_input() {
-                let t = results.get(idx).ok_or(GraphError::MissingResults)?[0].clone();
-                inputs.push(t);
-            } else {
-                for (idx, outlet) in n.inputs().iter() {
-                    match results.get(&idx) {
-                        Some(value) => inputs.push(value[*outlet].clone()),
-                        None => return Err(Box::new(GraphError::MissingNode(*idx))),
-                    }
-                }
-            };
-
-            debug!("executing {}: {}", idx, n.as_str());
-            debug!("dims: {:?}", n.out_dims());
-            debug!(
-                "input_dims: {:?}",
-                inputs.iter().map(|x| x.dims()).collect::<Vec<_>>()
-            );
-
-            debug!("input nodes: {:?}", n.inputs());
-
-            if n.is_lookup() {
-                let (mut min, mut max) = (0, 0);
-                for i in &inputs {
-                    max = max.max(
-                        i.iter()
-                            .map(|x| felt_to_i128(*x))
-                            .max()
-                            .ok_or("missing max")?,
-                    );
-                    min = min.min(
-                        i.iter()
-                            .map(|x| felt_to_i128(*x))
-                            .min()
-                            .ok_or("missing min")?,
-                    );
-                }
-                max_lookup_inputs = max_lookup_inputs.max(max);
-                min_lookup_inputs = min_lookup_inputs.min(min);
-                debug!("max lookup inputs: {}", max);
-                debug!("min lookup inputs: {}", min);
-            }
-
-            match n {
-                NodeType::Node(n) => {
-                    // execute the op
-                    let start = instant::Instant::now();
-                    let mut res = Op::<Fp>::f(&n.opkind, &inputs)?;
-                    res.output.reshape(&n.out_dims)?;
-                    let elapsed = start.elapsed();
-                    trace!("op took: {:?}", elapsed);
-                    // see if any of the intermediate lookup calcs are the max
-                    if !res.intermediate_lookups.is_empty() {
-                        let (mut min, mut max) = (0, 0);
-                        for i in &res.intermediate_lookups {
-                            max = max.max(i.clone().into_iter().max().ok_or("missing max")?);
-                            min = min.min(i.clone().into_iter().min().ok_or("missing min")?);
-                        }
-                        max_lookup_inputs = max_lookup_inputs.max(max);
-                        min_lookup_inputs = min_lookup_inputs.min(min);
-                        debug!("intermediate max lookup inputs: {}", max);
-                        debug!("intermediate min lookup inputs: {}", min);
-                    }
-                    debug!(
-                        "------------ output node int {}: {} \n ------------ float: {} \n ------------ max: {} \n ------------ min: {} \n ------------ scale: {}",
-                        idx,
-                        res.output.map(crate::fieldutils::felt_to_i32).show(),
-                        res.output
-                            .map(|x| crate::fieldutils::felt_to_f64(x)
-                                / scale_to_multiplier(n.out_scale))
-                            .show(),
-                        res.output.clone().into_iter().map(crate::fieldutils::felt_to_i128).max().unwrap_or(0),
-                        res.output.clone().into_iter().map(crate::fieldutils::felt_to_i128).min().unwrap_or(0),
-                        n.out_scale
-                    );
-                    results.insert(idx, vec![res.output]);
-                }
-                NodeType::SubGraph {
-                    model,
-                    output_mappings,
-                    input_mappings,
-                    inputs: input_tuple,
-                    ..
-                } => {
-                    let orig_inputs = inputs.clone();
-                    let input_mappings = input_mappings.clone();
-
-                    let input_dims = inputs.iter().map(|inp| inp.dims());
-                    let num_iter = number_of_iterations(&input_mappings, input_dims.collect());
-
-                    debug!(
-                        "{} iteration(s) in a subgraph with inputs {:?} and sources {:?}",
-                        num_iter, input_tuple, model.graph.inputs
-                    );
-
-                    debug!("input_mappings: {:?}", input_mappings);
-
-                    let mut full_results: Vec<Tensor<Fp>> = vec![];
-
-                    for i in 0..num_iter {
-                        // replace the Stacked input with the current chunk iter
-                        for ((mapping, inp), og_input) in
-                            input_mappings.iter().zip(&mut inputs).zip(&orig_inputs)
-                        {
-                            if let InputMapping::Stacked { axis, chunk } = mapping {
-                                let start = i * chunk;
-                                let end = (i + 1) * chunk;
-                                let t = crate::tensor::ops::slice(og_input, axis, &start, &end)?;
-                                *inp = t;
-                            }
-                        }
-
-                        let res = model.forward(&inputs)?;
-                        // recursively get the max lookup inputs for subgraphs
-                        max_lookup_inputs = max_lookup_inputs.max(res.max_lookup_inputs);
-                        min_lookup_inputs = min_lookup_inputs.min(res.min_lookup_inputs);
-
-                        let mut outlets = BTreeMap::new();
-                        for (mappings, outlet_res) in output_mappings.iter().zip(res.outputs) {
-                            for mapping in mappings {
-                                match mapping {
-                                    OutputMapping::Single { outlet, .. } => {
-                                        outlets.insert(outlet, outlet_res.clone());
-                                    }
-                                    OutputMapping::Stacked { outlet, axis, .. } => {
-                                        if !full_results.is_empty() {
-                                            let stacked_res = crate::tensor::ops::concat(
-                                                &[&full_results[*outlet], &outlet_res],
-                                                *axis,
-                                            )?;
-
-                                            outlets.insert(outlet, stacked_res);
-                                        } else {
-                                            outlets.insert(outlet, outlet_res.clone());
-                                        }
-                                    }
-                                }
-                            }
-                        }
-
-                        full_results = outlets.into_values().collect_vec();
-
-                        let output_states = output_state_idx(output_mappings);
-                        let input_states = input_state_idx(&input_mappings);
-
-                        assert_eq!(input_states.len(), output_states.len());
-
-                        for (input_idx, output_idx) in input_states.iter().zip(output_states) {
-                            inputs[*input_idx] = full_results[output_idx].clone();
-                        }
-                    }
-
-                    trace!(
-                        "------------ output subgraph node {}: {:?}",
-                        idx,
-                        full_results
-                            .iter()
-                            .map(|x|
-                            // convert to tensor i32
-                            x.map(crate::fieldutils::felt_to_i32).show())
-                            .collect_vec()
-                    );
-
-                    results.insert(idx, full_results);
-                }
-            }
-        }
-
-        let output_nodes = self.graph.outputs.iter();
-        debug!(
-            "model outputs are nodes: {:?}",
-            output_nodes.clone().collect_vec()
-        );
-        let outputs = output_nodes
-            .map(|(idx, outlet)| {
-                Ok(results.get(&idx).ok_or(GraphError::MissingResults)?[*outlet].clone())
-            })
-            .collect::<Result<Vec<_>, GraphError>>()?;
-
-        let res = ForwardResult {
-            outputs,
-            max_lookup_inputs,
-            min_lookup_inputs,
-        };
-
-        Ok(res)
+        let valtensor_inputs: Vec<ValTensor<Fp>> = model_inputs
+            .iter()
+            .map(|x| x.map(|elem| ValType::Value(Value::known(elem))).into())
+            .collect();
+        let res = self.dummy_layout(&RunArgs::default(), &valtensor_inputs)?;
+        Ok(res.into())
     }
 
     /// Loads an Onnx model from a specified path.
@@ -772,7 +611,7 @@ impl Model {
         for (symbol, value) in run_args.variables.iter() {
             let symbol = model.symbol_table.sym(symbol);
             symbol_values = symbol_values.with(&symbol, *value as i64);
-            info!("set {} to {}", symbol, value);
+            debug!("set {} to {}", symbol, value);
         }
 
         // Note: do not optimize the model, as the layout will depend on underlying hardware
@@ -1293,7 +1132,7 @@ impl Model {
 
         // Then number of columns in the circuits
         #[cfg(not(target_arch = "wasm32"))]
-        info!(
+        debug!(
             "{} {} {} (coord={}, constants={})",
             "model uses".blue(),
             num_rows.to_string().blue(),
@@ -1335,18 +1174,29 @@ impl Model {
             };
 
             debug!(
-                "laying out {}: {}, row:{}, coord:{}, total_constants: {}",
+                "laying out {}: {}, row:{}, coord:{}, total_constants: {}, max_lookup_inputs: {}, min_lookup_inputs: {}",
                 idx,
                 node.as_str(),
                 region.row(),
                 region.linear_coord(),
-                region.total_constants()
+                region.total_constants(),
+                region.max_lookup_inputs(),
+                region.min_lookup_inputs()
             );
             debug!("dims: {:?}", node.out_dims());
             debug!(
                 "input_dims {:?}",
                 values.iter().map(|v| v.dims()).collect_vec()
             );
+            debug!("output scales: {:?}", node.out_scales());
+            debug!("input indices: {:?}", node.inputs());
+            debug!(
+                "input scales: {:?}",
+                node.inputs()
+                    .iter()
+                    .map(|(idx, outlet)| self.graph.nodes[idx].out_scales()[*outlet])
+                    .collect_vec()
+            );
 
             match &node {
                 NodeType::Node(n) => {
@@ -1489,28 +1339,13 @@ impl Model {
     pub fn dummy_layout(
         &self,
         run_args: &RunArgs,
-        input_shapes: &[Vec<usize>],
+        inputs: &[ValTensor<Fp>],
     ) -> Result<DummyPassRes, Box<dyn Error>> {
-        info!("calculating num of constraints using dummy model layout...");
+        debug!("calculating num of constraints using dummy model layout...");
 
         let start_time = instant::Instant::now();
 
         let mut results = BTreeMap::<usize, Vec<ValTensor<Fp>>>::new();
-        let default_value = if !self.visibility.input.is_fixed() {
-            ValType::Value(Value::<Fp>::unknown())
-        } else {
-            ValType::Constant(Fp::ONE)
-        };
-
-        let inputs: Vec<ValTensor<Fp>> = input_shapes
-            .iter()
-            .map(|shape| {
-                let mut t: ValTensor<Fp> =
-                    vec![default_value.clone(); shape.iter().product()].into();
-                t.reshape(shape)?;
-                Ok(t)
-            })
-            .collect::<Result<Vec<_>, Box<dyn Error>>>()?;
 
         for (i, input_idx) in self.graph.inputs.iter().enumerate() {
             results.insert(*input_idx, vec![inputs[i].clone()]);
@@ -1545,12 +1380,12 @@ impl Model {
                 .collect::<Result<Vec<_>, Box<dyn Error>>>()?;
 
             let _ = outputs
-                .into_iter()
+                .iter()
                 .zip(comparator)
                 .map(|(o, c)| {
                     dummy_config.layout(
                         &mut region,
-                        &[o, c],
+                        &[o.clone(), c],
                         Box::new(HybridOp::RangeCheck(run_args.tolerance)),
                     )
                 })
@@ -1566,7 +1401,7 @@ impl Model {
 
         // Then number of columns in the circuits
         #[cfg(not(target_arch = "wasm32"))]
-        info!(
+        debug!(
             "{} {} {} (coord={}, constants={})",
             "model uses".blue(),
             region.row().to_string().blue(),
@@ -1575,12 +1410,23 @@ impl Model {
             region.total_constants().to_string().red()
         );
 
+        let outputs = outputs
+            .iter()
+            .map(|x| {
+                x.get_felt_evals()
+                    .unwrap_or(Tensor::new(Some(&[Fp::ZERO]), &[1]).unwrap())
+            })
+            .collect();
+
         let res = DummyPassRes {
             num_rows: region.row(),
             linear_coord: region.linear_coord(),
             total_const_size: region.total_constants(),
             lookup_ops: region.used_lookups(),
             range_checks: region.used_range_checks(),
+            max_lookup_inputs: region.max_lookup_inputs(),
+            min_lookup_inputs: region.min_lookup_inputs(),
+            outputs,
         };
 
         Ok(res)

src/graph/node.rs

@@ -219,8 +219,6 @@ impl Op<Fp> for RebaseScale {
         let mut res = Op::<Fp>::f(&*self.inner, x)?;
         let rebase_res = Op::<Fp>::f(&self.rebase_op, &[res.output])?;
         res.output = rebase_res.output;
-        res.intermediate_lookups
-            .extend(rebase_res.intermediate_lookups);
 
         Ok(res)
     }

src/graph/utilities.rs

@@ -439,17 +439,16 @@ pub fn new_op_from_onnx(
             let op = load_op::<ScatterElements>(node.op(), idx, node.op().name().to_string())?;
             let axis = op.axis;
 
-            let mut op =
-                SupportedOp::Hybrid(crate::circuit::ops::hybrid::HybridOp::ScatterElements {
-                    dim: axis,
-                    constant_idx: None,
-                });
+            let mut op = SupportedOp::Linear(crate::circuit::ops::poly::PolyOp::ScatterElements {
+                dim: axis,
+                constant_idx: None,
+            });
 
             // if param_visibility.is_public() {
             if let Some(c) = inputs[1].opkind().get_mutable_constant() {
                 inputs[1].decrement_use();
                 deleted_indices.push(1);
-                op = SupportedOp::Hybrid(crate::circuit::ops::hybrid::HybridOp::ScatterElements {
+                op = SupportedOp::Linear(crate::circuit::ops::poly::PolyOp::ScatterElements {
                     dim: axis,
                     constant_idx: Some(c.raw_values.map(|x| x as usize)),
                 })
@@ -478,17 +477,16 @@ pub fn new_op_from_onnx(
             let op = load_op::<GatherElements>(node.op(), idx, node.op().name().to_string())?;
             let axis = op.axis;
 
-            let mut op =
-                SupportedOp::Hybrid(crate::circuit::ops::hybrid::HybridOp::GatherElements {
-                    dim: axis,
-                    constant_idx: None,
-                });
+            let mut op = SupportedOp::Linear(crate::circuit::ops::poly::PolyOp::GatherElements {
+                dim: axis,
+                constant_idx: None,
+            });
 
             // if param_visibility.is_public() {
             if let Some(c) = inputs[1].opkind().get_mutable_constant() {
                 inputs[1].decrement_use();
                 deleted_indices.push(inputs.len() - 1);
-                op = SupportedOp::Hybrid(crate::circuit::ops::hybrid::HybridOp::GatherElements {
+                op = SupportedOp::Linear(crate::circuit::ops::poly::PolyOp::GatherElements {
                     dim: axis,
                     constant_idx: Some(c.raw_values.map(|x| x as usize)),
                 })

src/graph/vars.rs

@@ -410,7 +410,7 @@ impl<F: PrimeField + TensorType + PartialOrd> ModelVars<F> {
         num_constants: usize,
         module_requires_fixed: bool,
     ) -> Self {
-        info!("number of blinding factors: {}", cs.blinding_factors());
+        debug!("number of blinding factors: {}", cs.blinding_factors());
 
         let advices = (0..3)
             .map(|_| VarTensor::new_advice(cs, logrows, num_inner_cols, var_len))

src/lib.rs

@@ -7,7 +7,6 @@
     overflowing_literals,
     path_statements,
     patterns_in_fns_without_body,
-    private_in_public,
     unconditional_recursion,
     unused,
     unused_allocation,
@@ -71,9 +70,32 @@ pub mod tensor;
 #[cfg(all(target_arch = "wasm32", target_os = "unknown"))]
 pub mod wasm;
 
+#[cfg(not(target_arch = "wasm32"))]
+use lazy_static::lazy_static;
+
 /// The denominator in the fixed point representation used when quantizing inputs
 pub type Scale = i32;
 
+#[cfg(not(target_arch = "wasm32"))]
+// Buf writer capacity
+lazy_static! {
+    /// The capacity of the buffer used for writing to disk
+    pub static ref EZKL_BUF_CAPACITY: usize = std::env::var("EZKL_BUF_CAPACITY")
+        .unwrap_or("8000".to_string())
+        .parse()
+        .unwrap();
+
+    /// The serialization format for the keys
+    pub static ref EZKL_KEY_FORMAT: String = std::env::var("EZKL_KEY_FORMAT")
+        .unwrap_or("raw-bytes".to_string());
+}
+
+#[cfg(target_arch = "wasm32")]
+const EZKL_KEY_FORMAT: &str = "raw-bytes";
+
+#[cfg(target_arch = "wasm32")]
+const EZKL_BUF_CAPACITY: &usize = &8000;
+
 /// Parameters specific to a proving run
 #[derive(Debug, Args, Deserialize, Serialize, Clone, PartialEq, PartialOrd)]
 pub struct RunArgs {

src/pfsys/mod.rs

@@ -8,6 +8,7 @@ use crate::circuit::CheckMode;
 use crate::graph::GraphWitness;
 use crate::pfsys::evm::aggregation::PoseidonTranscript;
 use crate::tensor::TensorType;
+use crate::{EZKL_BUF_CAPACITY, EZKL_KEY_FORMAT};
 use clap::ValueEnum;
 use halo2_proofs::circuit::Value;
 use halo2_proofs::plonk::{
@@ -42,6 +43,15 @@ use thiserror::Error as thisError;
 
 use halo2curves::bn256::{Bn256, Fr, G1Affine};
 
+fn serde_format_from_str(s: &str) -> halo2_proofs::SerdeFormat {
+    match s {
+        "processed" => halo2_proofs::SerdeFormat::Processed,
+        "raw-bytes-unchecked" => halo2_proofs::SerdeFormat::RawBytesUnchecked,
+        "raw-bytes" => halo2_proofs::SerdeFormat::RawBytes,
+        _ => panic!("invalid serde format"),
+    }
+}
+
 #[allow(missing_docs)]
 #[derive(
     ValueEnum, Copy, Clone, Default, Debug, PartialEq, Eq, Deserialize, Serialize, PartialOrd,
@@ -315,7 +325,7 @@ where
     /// Saves the Proof to a specified `proof_path`.
     pub fn save(&self, proof_path: &PathBuf) -> Result<(), Box<dyn Error>> {
         let file = std::fs::File::create(proof_path)?;
-        let mut writer = BufWriter::new(file);
+        let mut writer = BufWriter::with_capacity(*EZKL_BUF_CAPACITY, file);
         serde_json::to_writer(&mut writer, &self)?;
         Ok(())
     }
@@ -328,8 +338,10 @@ where
         <C as CurveAffine>::ScalarExt: FromUniformBytes<64>,
     {
         trace!("reading proof");
-        let data = std::fs::read_to_string(proof_path)?;
-        serde_json::from_str(&data).map_err(|e| e.into())
+        let file = std::fs::File::open(proof_path)?;
+        let reader = BufReader::with_capacity(*EZKL_BUF_CAPACITY, file);
+        let proof: Self = serde_json::from_reader(reader)?;
+        Ok(proof)
     }
 }
 
@@ -555,6 +567,7 @@ where
             verifier_params,
             pk.get_vk(),
             strategy,
+            verifier_params.n(),
         )?;
     }
     let elapsed = now.elapsed();
@@ -642,6 +655,7 @@ pub fn verify_proof_circuit<
     params: &'params Scheme::ParamsVerifier,
     vk: &VerifyingKey<Scheme::Curve>,
     strategy: Strategy,
+    orig_n: u64,
 ) -> Result<Strategy::Output, halo2_proofs::plonk::Error>
 where
     Scheme::Scalar: SerdeObject
@@ -662,7 +676,7 @@ where
     trace!("instances {:?}", instances);
 
     let mut transcript = TranscriptReadBuffer::init(Cursor::new(snark.proof.clone()));
-    verify_proof::<Scheme, V, _, TR, _>(params, vk, strategy, instances, &mut transcript)
+    verify_proof::<Scheme, V, _, TR, _>(params, vk, strategy, instances, &mut transcript, orig_n)
 }
 
 /// Loads a [VerifyingKey] at `path`.
@@ -678,10 +692,10 @@ where
     info!("loading verification key from {:?}", path);
     let f =
         File::open(path.clone()).map_err(|_| format!("failed to load vk at {}", path.display()))?;
-    let mut reader = BufReader::new(f);
+    let mut reader = BufReader::with_capacity(*EZKL_BUF_CAPACITY, f);
     VerifyingKey::<Scheme::Curve>::read::<_, C>(
         &mut reader,
-        halo2_proofs::SerdeFormat::RawBytes,
+        serde_format_from_str(&EZKL_KEY_FORMAT),
         params,
     )
     .map_err(Box::<dyn Error>::from)
@@ -700,10 +714,10 @@ where
     info!("loading proving key from {:?}", path);
     let f =
         File::open(path.clone()).map_err(|_| format!("failed to load pk at {}", path.display()))?;
-    let mut reader = BufReader::new(f);
+    let mut reader = BufReader::with_capacity(*EZKL_BUF_CAPACITY, f);
     ProvingKey::<Scheme::Curve>::read::<_, C>(
         &mut reader,
-        halo2_proofs::SerdeFormat::RawBytes,
+        serde_format_from_str(&EZKL_KEY_FORMAT),
         params,
     )
     .map_err(Box::<dyn Error>::from)
@@ -712,7 +726,7 @@ where
 /// Saves a [ProvingKey] to `path`.
 pub fn save_pk<Scheme: CommitmentScheme>(
     path: &PathBuf,
-    vk: &ProvingKey<Scheme::Curve>,
+    pk: &ProvingKey<Scheme::Curve>,
 ) -> Result<(), io::Error>
 where
     Scheme::Curve: SerdeObject + CurveAffine,
@@ -720,8 +734,8 @@ where
 {
     info!("saving proving key 💾");
     let f = File::create(path)?;
-    let mut writer = BufWriter::new(f);
-    vk.write(&mut writer, halo2_proofs::SerdeFormat::RawBytes)?;
+    let mut writer = BufWriter::with_capacity(*EZKL_BUF_CAPACITY, f);
+    pk.write(&mut writer, serde_format_from_str(&EZKL_KEY_FORMAT))?;
     writer.flush()?;
     Ok(())
 }
@@ -737,8 +751,8 @@ where
 {
     info!("saving verification key 💾");
     let f = File::create(path)?;
-    let mut writer = BufWriter::new(f);
-    vk.write(&mut writer, halo2_proofs::SerdeFormat::RawBytes)?;
+    let mut writer = BufWriter::with_capacity(*EZKL_BUF_CAPACITY, f);
+    vk.write(&mut writer, serde_format_from_str(&EZKL_KEY_FORMAT))?;
     writer.flush()?;
     Ok(())
 }
@@ -750,7 +764,7 @@ pub fn save_params<Scheme: CommitmentScheme>(
 ) -> Result<(), io::Error> {
     info!("saving parameters 💾");
     let f = File::create(path)?;
-    let mut writer = BufWriter::new(f);
+    let mut writer = BufWriter::with_capacity(*EZKL_BUF_CAPACITY, f);
     params.write(&mut writer)?;
     writer.flush()?;
     Ok(())
@@ -840,6 +854,7 @@ pub(crate) fn verify_proof_circuit_kzg<
     proof: Snark<Fr, G1Affine>,
     vk: &VerifyingKey<G1Affine>,
     strategy: Strategy,
+    orig_n: u64,
 ) -> Result<Strategy::Output, halo2_proofs::plonk::Error> {
     match proof.transcript_type {
         TranscriptType::EVM => verify_proof_circuit::<
@@ -849,7 +864,7 @@ pub(crate) fn verify_proof_circuit_kzg<
             _,
             _,
             EvmTranscript<G1Affine, _, _, _>,
-        >(&proof, params, vk, strategy),
+        >(&proof, params, vk, strategy, orig_n),
         TranscriptType::Poseidon => verify_proof_circuit::<
             Fr,
             VerifierSHPLONK<'_, Bn256>,
@@ -857,7 +872,7 @@ pub(crate) fn verify_proof_circuit_kzg<
             _,
             _,
             PoseidonTranscript<NativeLoader, _>,
-        >(&proof, params, vk, strategy),
+        >(&proof, params, vk, strategy, orig_n),
     }
 }
 

src/python.rs

@@ -15,7 +15,7 @@ use crate::graph::{
 use crate::pfsys::evm::aggregation::AggregationCircuit;
 use crate::pfsys::{
     load_pk, load_vk, save_params, save_vk, srs::gen_srs as ezkl_gen_srs, srs::load_srs, ProofType,
-    Snark, TranscriptType,
+    TranscriptType,
 };
 use crate::RunArgs;
 use ethers::types::H160;
@@ -689,14 +689,23 @@ fn prove(
     settings_path=PathBuf::from(DEFAULT_SETTINGS),
     vk_path=PathBuf::from(DEFAULT_VK),
     srs_path=None,
+    non_reduced_srs=Some(DEFAULT_USE_REDUCED_SRS_FOR_VERIFICATION.parse::<bool>().unwrap()),
 ))]
 fn verify(
     proof_path: PathBuf,
     settings_path: PathBuf,
     vk_path: PathBuf,
     srs_path: Option<PathBuf>,
+    non_reduced_srs: Option<bool>,
 ) -> Result<bool, PyErr> {
-    crate::execute::verify(proof_path, settings_path, vk_path, srs_path).map_err(|e| {
+    crate::execute::verify(
+        proof_path,
+        settings_path,
+        vk_path,
+        srs_path,
+        non_reduced_srs,
+    )
+    .map_err(|e| {
         let err_str = format!("Failed to run verify: {}", e);
         PyRuntimeError::new_err(err_str)
     })?;
@@ -1097,16 +1106,6 @@ fn create_evm_verifier_aggr(
     Ok(true)
 }
 
-/// print hex representation of a proof
-#[pyfunction(signature = (proof_path))]
-fn print_proof_hex(proof_path: PathBuf) -> Result<String, PyErr> {
-    let proof = Snark::load::<KZGCommitmentScheme<Bn256>>(&proof_path)
-        .map_err(|_| PyIOError::new_err("Failed to load proof"))?;
-
-    let hex_str = hex::encode(proof.proof);
-    Ok(format!("0x{}", hex_str))
-}
-
 // Python Module
 #[pymodule]
 fn ezkl(_py: Python<'_>, m: &PyModule) -> PyResult<()> {
@@ -1145,7 +1144,6 @@ fn ezkl(_py: Python<'_>, m: &PyModule) -> PyResult<()> {
     m.add_function(wrap_pyfunction!(deploy_vk_evm, m)?)?;
     m.add_function(wrap_pyfunction!(deploy_da_evm, m)?)?;
     m.add_function(wrap_pyfunction!(verify_evm, m)?)?;
-    m.add_function(wrap_pyfunction!(print_proof_hex, m)?)?;
     m.add_function(wrap_pyfunction!(setup_test_evm_witness, m)?)?;
     m.add_function(wrap_pyfunction!(create_evm_verifier_aggr, m)?)?;
     m.add_function(wrap_pyfunction!(create_evm_data_attestation, m)?)?;

src/tensor/mod.rs

@@ -580,16 +580,16 @@ impl<T: Clone + TensorType> Tensor<T> {
     /// use ezkl::tensor::Tensor;
     /// let mut a = Tensor::<i32>::new(Some(&[1,2,3,4,5,6]), &[2, 3]).unwrap();
     /// let expected = Tensor::<i32>::new(Some(&[1, 2, 3, 4, 5, 6, 0, 0]), &[8]).unwrap();
-    /// assert_eq!(a.pad_to_zero_rem(4).unwrap(), expected);
+    /// assert_eq!(a.pad_to_zero_rem(4, 0).unwrap(), expected);
     ///
     /// let expected = Tensor::<i32>::new(Some(&[1, 2, 3, 4, 5, 6, 0, 0, 0]), &[9]).unwrap();
-    /// assert_eq!(a.pad_to_zero_rem(9).unwrap(), expected);
+    /// assert_eq!(a.pad_to_zero_rem(9, 0).unwrap(), expected);
     /// ```
-    pub fn pad_to_zero_rem(&self, n: usize) -> Result<Tensor<T>, TensorError> {
+    pub fn pad_to_zero_rem(&self, n: usize, pad: T) -> Result<Tensor<T>, TensorError> {
         let mut inner = self.inner.clone();
         let remainder = self.len() % n;
         if remainder != 0 {
-            inner.resize(self.len() + n - remainder, T::zero().unwrap());
+            inner.resize(self.len() + n - remainder, pad);
         }
         Tensor::new(Some(&inner), &[inner.len()])
     }

src/tensor/ops.rs

@@ -243,7 +243,7 @@ pub fn and<
 /// Some(&[1, 0, 1, 0, 1, 0]),
 /// &[2, 3],
 /// ).unwrap();
-/// let result = equals(&a, &b).unwrap().0;
+/// let result = equals(&a, &b).unwrap();
 /// let expected = Tensor::<i128>::new(Some(&[1, 0, 1, 0, 1, 1]), &[2, 3]).unwrap();
 /// assert_eq!(result, expected);
 /// ```
@@ -260,7 +260,7 @@ pub fn equals<
 >(
     a: &Tensor<T>,
     b: &Tensor<T>,
-) -> Result<(Tensor<T>, Vec<Tensor<T>>), TensorError> {
+) -> Result<Tensor<T>, TensorError> {
     let a = a.clone();
     let b = b.clone();
 
@@ -268,7 +268,7 @@ pub fn equals<
 
     let result = nonlinearities::kronecker_delta(&diff);
 
-    Ok((result, vec![diff]))
+    Ok(result)
 }
 
 /// Greater than operation.
@@ -289,7 +289,7 @@ pub fn equals<
 /// ).unwrap();
 /// let result = greater(&a, &b).unwrap();
 /// let expected = Tensor::<i128>::new(Some(&[0, 1, 1, 0, 0, 0]), &[2, 3]).unwrap();
-/// assert_eq!(result.0, expected);
+/// assert_eq!(result, expected);
 /// ```
 pub fn greater<
     T: TensorType
@@ -302,7 +302,7 @@ pub fn greater<
 >(
     a: &Tensor<T>,
     b: &Tensor<T>,
-) -> Result<(Tensor<T>, Vec<Tensor<T>>), TensorError> {
+) -> Result<Tensor<T>, TensorError> {
     let mask_inter = (a.clone() - b.clone())?;
     let mask = mask_inter.map(|x| {
         if x > T::zero().ok_or(TensorError::Unsupported).unwrap() {
@@ -311,7 +311,7 @@ pub fn greater<
             T::zero().ok_or(TensorError::Unsupported).unwrap()
         }
     });
-    Ok((mask, vec![mask_inter]))
+    Ok(mask)
 }
 
 /// Greater equals than operation.
@@ -332,7 +332,7 @@ pub fn greater<
 /// ).unwrap();
 /// let result = greater_equal(&a, &b).unwrap();
 /// let expected = Tensor::<i128>::new(Some(&[1, 1, 1, 1, 0, 0]), &[2, 3]).unwrap();
-/// assert_eq!(result.0, expected);
+/// assert_eq!(result, expected);
 /// ```
 pub fn greater_equal<
     T: TensorType
@@ -345,7 +345,7 @@ pub fn greater_equal<
 >(
     a: &Tensor<T>,
     b: &Tensor<T>,
-) -> Result<(Tensor<T>, Vec<Tensor<T>>), TensorError> {
+) -> Result<Tensor<T>, TensorError> {
     let mask_inter = (a.clone() - b.clone())?;
     let mask = mask_inter.map(|x| {
         if x >= T::zero().ok_or(TensorError::Unsupported).unwrap() {
@@ -354,7 +354,7 @@ pub fn greater_equal<
             T::zero().ok_or(TensorError::Unsupported).unwrap()
         }
     });
-    Ok((mask, vec![mask_inter]))
+    Ok(mask)
 }
 
 /// Less than to operation.
@@ -375,7 +375,7 @@ pub fn greater_equal<
 /// ).unwrap();
 /// let result = less(&a, &b).unwrap();
 /// let expected = Tensor::<i128>::new(Some(&[0, 1, 0, 0, 0, 1]), &[2, 3]).unwrap();
-/// assert_eq!(result.0, expected);
+/// assert_eq!(result, expected);
 /// ```
 ///
 pub fn less<
@@ -389,7 +389,7 @@ pub fn less<
 >(
     a: &Tensor<T>,
     b: &Tensor<T>,
-) -> Result<(Tensor<T>, Vec<Tensor<T>>), TensorError> {
+) -> Result<Tensor<T>, TensorError> {
     // a < b <=> b > a
     greater(b, a)
 }
@@ -412,7 +412,7 @@ pub fn less<
 /// ).unwrap();
 /// let result = less_equal(&a, &b).unwrap();
 /// let expected = Tensor::<i128>::new(Some(&[1, 1, 0, 1, 1, 1]), &[2, 3]).unwrap();
-/// assert_eq!(result.0, expected);
+/// assert_eq!(result, expected);
 /// ```
 ///
 pub fn less_equal<
@@ -426,7 +426,7 @@ pub fn less_equal<
 >(
     a: &Tensor<T>,
     b: &Tensor<T>,
-) -> Result<(Tensor<T>, Vec<Tensor<T>>), TensorError> {
+) -> Result<Tensor<T>, TensorError> {
     // a < b <=> b > a
     greater_equal(b, a)
 }
@@ -2300,12 +2300,12 @@ pub fn deconv<
 ///     Some(&[5, 2, 3, 0, 4, -1, 3, 1, 6]),
 ///     &[1, 1, 3, 3],
 /// ).unwrap();
-/// let pooled = sumpool(&x, [(0, 0); 2], (1, 1), (2, 2), false).unwrap().0;
+/// let pooled = sumpool(&x, [(0, 0); 2], (1, 1), (2, 2), false).unwrap();
 /// let expected: Tensor<i128> = Tensor::<i128>::new(Some(&[11, 8, 8, 10]), &[1, 1, 2, 2]).unwrap();
 /// assert_eq!(pooled, expected);
 ///
 /// // This time with normalization
-/// let pooled = sumpool(&x, [(0, 0); 2], (1, 1), (2, 2), true).unwrap().0;
+/// let pooled = sumpool(&x, [(0, 0); 2], (1, 1), (2, 2), true).unwrap();
 /// let expected: Tensor<i128> = Tensor::<i128>::new(Some(&[3, 2, 2, 3]), &[1, 1, 2, 2]).unwrap();
 /// assert_eq!(pooled, expected);
 /// ```
@@ -2315,7 +2315,7 @@ pub fn sumpool(
     stride: (usize, usize),
     kernel_shape: (usize, usize),
     normalize: bool,
-) -> Result<(Tensor<i128>, Vec<Tensor<i128>>), TensorError> {
+) -> Result<Tensor<i128>, TensorError> {
     let image_dims = image.dims();
     let batch_size = image_dims[0];
     let image_channels = image_dims[1];
@@ -2345,15 +2345,12 @@ pub fn sumpool(
     let mut combined = res.combine()?;
     combined.reshape(&[&[batch_size, image_channels], shape].concat())?;
 
-    let mut inter = vec![];
-
     if normalize {
-        inter.push(combined.clone());
         let norm = kernel.len();
         combined = nonlinearities::const_div(&combined, norm as f64);
     }
 
-    Ok((combined, inter))
+    Ok(combined)
 }
 
 /// Applies 2D max pooling over a 4D tensor of shape B x C x H x W.
@@ -3048,11 +3045,7 @@ pub mod nonlinearities {
     }
 
     /// softmax layout
-    pub fn softmax_axes(
-        a: &Tensor<i128>,
-        scale: f64,
-        axes: &[usize],
-    ) -> (Tensor<i128>, Vec<Tensor<i128>>) {
+    pub fn softmax_axes(a: &Tensor<i128>, scale: f64, axes: &[usize]) -> Tensor<i128> {
         // we want this to be as small as possible so we set the output scale to 1
         let dims = a.dims();
 
@@ -3060,8 +3053,6 @@ pub mod nonlinearities {
             return softmax(a, scale);
         }
 
-        let mut intermediate_values = vec![];
-
         let cartesian_coord = dims[..dims.len() - 1]
             .iter()
             .map(|x| 0..*x)
@@ -3084,8 +3075,7 @@ pub mod nonlinearities {
 
             let res = softmax(&softmax_input, scale);
 
-            outputs.push(res.0);
-            intermediate_values.extend(res.1);
+            outputs.push(res);
         }
 
         let mut res = Tensor::new(Some(&outputs), &[outputs.len()])
@@ -3093,7 +3083,7 @@ pub mod nonlinearities {
             .combine()
             .unwrap();
         res.reshape(dims).unwrap();
-        (res, intermediate_values)
+        res
     }
 
     /// Applies softmax
@@ -3110,24 +3100,20 @@ pub mod nonlinearities {
     ///     Some(&[2, 2, 3, 2, 2, 0]),
     ///     &[2, 3],
     /// ).unwrap();
-    /// let result = softmax(&x, 128.0).0;
+    /// let result = softmax(&x, 128.0);
     /// // doubles the scale of the input
     /// let expected = Tensor::<i128>::new(Some(&[2730, 2730, 2751, 2730, 2730, 2688]), &[2, 3]).unwrap();
     /// assert_eq!(result, expected);
     /// ```
-    pub fn softmax(a: &Tensor<i128>, scale: f64) -> (Tensor<i128>, Vec<Tensor<i128>>) {
+    pub fn softmax(a: &Tensor<i128>, scale: f64) -> Tensor<i128> {
         // the more accurate calculation is commented out and we implement as below so it matches the steps in layout
-        let mut intermediate_values = vec![];
-
-        intermediate_values.push(a.clone());
 
         let exp = exp(a, scale);
 
         let sum = sum(&exp).unwrap();
-        intermediate_values.push(sum.clone());
         let inv_denom = recip(&sum, scale, scale);
 
-        ((exp * inv_denom).unwrap(), intermediate_values)
+        (exp * inv_denom).unwrap()
     }
 
     /// Applies range_check_percent

src/tensor/val.rs

@@ -454,12 +454,12 @@ impl<F: PrimeField + TensorType + PartialOrd> ValTensor<F> {
     }
 
     /// Calls `pad_to_zero_rem` on the inner tensor.
-    pub fn pad_to_zero_rem(&mut self, n: usize) -> Result<(), Box<dyn Error>> {
+    pub fn pad_to_zero_rem(&mut self, n: usize, pad: ValType<F>) -> Result<(), Box<dyn Error>> {
         match self {
             ValTensor::Value {
                 inner: v, dims: d, ..
             } => {
-                *v = v.pad_to_zero_rem(n)?;
+                *v = v.pad_to_zero_rem(n, pad)?;
                 *d = v.dims().to_vec();
             }
             ValTensor::Instance { .. } => {

src/wasm.rs

@@ -311,13 +311,19 @@ pub fn verify(
     let vk = VerifyingKey::<G1Affine>::read::<_, GraphCircuit>(
         &mut reader,
         halo2_proofs::SerdeFormat::RawBytes,
-        circuit_settings,
+        circuit_settings.clone(),
     )
     .map_err(|e| JsError::new(&format!("Failed to deserialize vk: {}", e)))?;
 
     let strategy = KZGSingleStrategy::new(params.verifier_params());
 
-    let result = verify_proof_circuit_kzg(params.verifier_params(), snark, &vk, strategy);
+    let result = verify_proof_circuit_kzg(
+        params.verifier_params(),
+        snark,
+        &vk,
+        strategy,
+        1 << circuit_settings.run_args.logrows,
+    );
 
     match result {
         Ok(_) => Ok(true),
@@ -387,15 +393,6 @@ pub fn prove(
         .into_bytes())
 }
 
-/// print hex representation of a proof
-#[wasm_bindgen]
-#[allow(non_snake_case)]
-pub fn printProofHex(proof: wasm_bindgen::Clamped<Vec<u8>>) -> Result<String, JsError> {
-    let proof: crate::pfsys::Snark<Fr, G1Affine> = serde_json::from_slice(&proof[..])
-        .map_err(|e| JsError::new(&format!("Failed to deserialize proof: {}", e)))?;
-    let hex_str = hex::encode(proof.proof);
-    Ok(format!("0x{}", hex_str))
-}
 // VALIDATION FUNCTIONS
 
 /// Witness file validation

tests/integration_tests.rs

@@ -560,7 +560,7 @@ mod native_tests {
                 crate::native_tests::setup_py_env();
                 let test_dir = TempDir::new(test).unwrap();
                 let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
-                accuracy_measurement(path, test.to_string(), "private", "private", "public", 1, "resources", 18.0, false);
+                accuracy_measurement(path, test.to_string(), "private", "private", "public", 1, "resources", 3.1, false);
                 test_dir.close().unwrap();
             }
 
@@ -1745,6 +1745,30 @@ mod native_tests {
             .status()
             .expect("failed to execute process");
         assert!(status.success());
+
+        // load settings file
+        let settings =
+            std::fs::read_to_string(settings_path.clone()).expect("failed to read settings file");
+
+        let graph_settings = serde_json::from_str::<GraphSettings>(&settings)
+            .expect("failed to parse settings file");
+
+        // get_srs for the graph_settings_num_instances
+        let _ = download_srs(graph_settings.log2_total_instances());
+
+        let status = Command::new(format!("{}/release/ezkl", *CARGO_TARGET_DIR))
+            .args([
+                "verify",
+                format!("--settings-path={}", settings_path).as_str(),
+                "--proof-path",
+                &format!("{}/{}/proof.pf", test_dir, example_name),
+                "--vk-path",
+                &format!("{}/{}/key.vk", test_dir, example_name),
+                "--reduced-srs=true",
+            ])
+            .status()
+            .expect("failed to execute process");
+        assert!(status.success());
     }
 
     // prove-serialize-verify, the usual full path

tests/py_integration_tests.rs

@@ -131,7 +131,7 @@ mod py_tests {
         "simple_demo_aggregated_proofs.ipynb",
         "ezkl_demo.ipynb", // 10
         "lstm.ipynb",
-        "set_membership.ipynb",
+        "set_membership.ipynb", // 12
         "decision_tree.ipynb",
         "random_forest.ipynb",
         "gradient_boosted_trees.ipynb", // 15

tests/python/binding_tests.py

@@ -113,13 +113,13 @@ def test_calibrate_over_user_range():
     data_path = os.path.join(
         examples_path,
         'onnx',
-        '1l_average',
+        '1l_relu',
         'input.json'
     )
     model_path = os.path.join(
         examples_path,
         'onnx',
-        '1l_average',
+        '1l_relu',
         'network.onnx'
     )
     output_path = os.path.join(
@@ -147,13 +147,13 @@ def test_calibrate():
     data_path = os.path.join(
         examples_path,
         'onnx',
-        '1l_average',
+        '1l_relu',
         'input.json'
     )
     model_path = os.path.join(
         examples_path,
         'onnx',
-        '1l_average',
+        '1l_relu',
         'network.onnx'
     )
     output_path = os.path.join(
@@ -183,7 +183,7 @@ def test_model_compile():
     model_path = os.path.join(
         examples_path,
         'onnx',
-        '1l_average',
+        '1l_relu',
         'network.onnx'
     )
     compiled_model_path = os.path.join(
@@ -205,7 +205,7 @@ def test_forward():
     data_path = os.path.join(
         examples_path,
         'onnx',
-        '1l_average',
+        '1l_relu',
         'input.json'
     )
     model_path = os.path.join(
@@ -392,9 +392,7 @@ def test_prove_evm():
     assert res['transcript_type'] == 'EVM'
     assert os.path.isfile(proof_path)
 
-    res = ezkl.print_proof_hex(proof_path)
-    # to figure out a better way of testing print_proof_hex
-    assert type(res) == str
+
 
 
 def test_create_evm_verifier():

tests/wasm.rs

@@ -9,8 +9,8 @@ mod wasm32 {
     use ezkl::pfsys;
     use ezkl::wasm::{
         bufferToVecOfstring, compiledCircuitValidation, encodeVerifierCalldata, genPk, genVk,
-        genWitness, inputValidation, pkValidation, poseidonHash, printProofHex, proofValidation,
-        prove, settingsValidation, srsValidation, stringToFelt, stringToFloat, stringToInt,
+        genWitness, inputValidation, pkValidation, poseidonHash, proofValidation, prove,
+        settingsValidation, srsValidation, stringToFelt, stringToFloat, stringToInt,
         u8_array_to_u128_le, verify, vkValidation, witnessValidation,
     };
     use halo2_solidity_verifier::encode_calldata;
@@ -258,15 +258,6 @@ mod wasm32 {
         assert!(value);
     }
 
-    #[wasm_bindgen_test]
-    async fn print_proof_hex_test() {
-        let proof = printProofHex(wasm_bindgen::Clamped(PROOF.to_vec()))
-            .map_err(|_| "failed")
-            .unwrap();
-
-        assert!(proof.len() > 0);
-    }
-
     #[wasm_bindgen_test]
     async fn verify_validations() {
         // Run witness validation on network (should fail)