Merge branch 'main' into example-reorg

.github/workflows/large-tests.yml

@@ -11,7 +11,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2024-01-04
+          toolchain: nightly-2023-08-24
           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-2024-01-04
+          toolchain: nightly-2023-08-24
           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-2024-01-04
+          toolchain: nightly-2023-08-24
           override: true
           components: rustfmt, clippy
       - name: Build
@@ -38,7 +38,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2024-01-04
+          toolchain: nightly-2023-08-24
           override: true
           components: rustfmt, clippy
       - name: Docs
@@ -50,7 +50,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2024-01-04
+          toolchain: nightly-2023-08-24
           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-2024-01-04
+          toolchain: nightly-2023-08-24
           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-2024-01-04
+          toolchain: nightly-2023-08-24
           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-2024-01-04
+          toolchain: nightly-2023-08-24
           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-2024-01-04
+          toolchain: nightly-2023-08-24
           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-2024-01-04
+          toolchain: nightly-2023-08-24
           override: true
           components: rustfmt, clippy
       - uses: jetli/wasm-pack-action@v0.4.0
@@ -198,8 +198,10 @@ jobs:
         #   chromedriver-version: "115.0.5790.102"
       - name: Install wasm32-unknown-unknown
         run: rustup target add wasm32-unknown-unknown
+      - name: Install wasm runner
+        run: cargo install wasm-server-runner
       - name: Add rust-src
-        run: rustup component add rust-src --toolchain nightly-2024-01-04-x86_64-unknown-linux-gnu
+        run: rustup component add rust-src --toolchain nightly-2023-08-24-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 +214,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2024-01-04
+          toolchain: nightly-2023-08-24
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -229,7 +231,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2024-01-04
+          toolchain: nightly-2023-08-24
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -284,7 +286,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2024-01-04
+          toolchain: nightly-2023-08-24
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -310,7 +312,7 @@ 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 b320f350156a0fb15c2eb13dc380deb2367c4474 --profile local --locked anvil --force
+        run: cargo install --git https://github.com/foundry-rs/foundry --rev 95a93cd397f25f3f8d49d2851eb52bc2d52dd983 --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)
@@ -343,15 +345,18 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2024-01-04
+          toolchain: nightly-2023-08-24
           override: true
           components: rustfmt, clippy
       - uses: jetli/wasm-pack-action@v0.4.0
       - name: Add wasm32-unknown-unknown target
         run: rustup target add wasm32-unknown-unknown
 
+      - name: Install wasm-server-runner
+        run: cargo install wasm-server-runner
+
       - name: Add rust-src
-        run: rustup component add rust-src --toolchain nightly-2024-01-04-x86_64-unknown-linux-gnu
+        run: rustup component add rust-src --toolchain nightly-2023-08-24-x86_64-unknown-linux-gnu
       - uses: actions/checkout@v3
       - name: Use pnpm 8
         uses: pnpm/action-setup@v2
@@ -411,11 +416,11 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2024-01-04
+          toolchain: nightly-2023-08-24
           override: true
           components: rustfmt, clippy
       - name: Add rust-src
-        run: rustup component add rust-src --toolchain nightly-2024-01-04-x86_64-unknown-linux-gnu
+        run: rustup component add rust-src --toolchain nightly-2023-08-24-x86_64-unknown-linux-gnu
       - uses: actions/checkout@v3
       - uses: baptiste0928/cargo-install@v1
         with:
@@ -445,7 +450,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2024-01-04
+          toolchain: nightly-2023-08-24
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -455,7 +460,7 @@ 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 b320f350156a0fb15c2eb13dc380deb2367c4474 --profile local --locked anvil --force
+        run: cargo install --git https://github.com/foundry-rs/foundry --rev 95a93cd397f25f3f8d49d2851eb52bc2d52dd983 --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 +473,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2024-01-04
+          toolchain: nightly-2023-08-24
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -486,7 +491,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2024-01-04
+          toolchain: nightly-2023-08-24
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -503,7 +508,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2024-01-04
+          toolchain: nightly-2023-08-24
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -520,7 +525,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2024-01-04
+          toolchain: nightly-2023-08-24
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -530,7 +535,7 @@ 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 b320f350156a0fb15c2eb13dc380deb2367c4474 --profile local --locked anvil --force
+        run: cargo install --git https://github.com/foundry-rs/foundry --rev 95a93cd397f25f3f8d49d2851eb52bc2d52dd983 --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 +546,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2024-01-04
+          toolchain: nightly-2023-08-24
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -563,7 +568,7 @@ jobs:
           python-version: "3.7"
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2024-01-04
+          toolchain: nightly-2023-08-24
           override: true
           components: rustfmt, clippy
       - name: Install solc
@@ -571,7 +576,7 @@ 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 b320f350156a0fb15c2eb13dc380deb2367c4474 --profile local --locked anvil --force
+        run: cargo install --git https://github.com/foundry-rs/foundry --rev 95a93cd397f25f3f8d49d2851eb52bc2d52dd983 --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 +592,7 @@ jobs:
           python-version: "3.7"
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2024-01-04
+          toolchain: nightly-2023-08-24
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -618,7 +623,7 @@ jobs:
           python-version: "3.9"
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2024-01-04
+          toolchain: nightly-2023-08-24
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -628,7 +633,7 @@ 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 b320f350156a0fb15c2eb13dc380deb2367c4474 --profile local --locked anvil --force
+        run: cargo install --git https://github.com/foundry-rs/foundry --rev 95a93cd397f25f3f8d49d2851eb52bc2d52dd983 --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

@@ -22,15 +22,18 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2024-01-04
+          toolchain: nightly-2023-08-24
           override: true
           components: rustfmt, clippy
       - uses: jetli/wasm-pack-action@v0.4.0
       - name: Add wasm32-unknown-unknown target
         run: rustup target add wasm32-unknown-unknown
 
+      - name: Install wasm-server-runner
+        run: cargo install wasm-server-runner
+
       - name: Add rust-src
-        run: rustup component add rust-src --toolchain nightly-2024-01-04-x86_64-unknown-linux-gnu
+        run: rustup component add rust-src --toolchain nightly-2023-08-24-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 = { git = "https://github.com/privacy-scaling-explorations/halo2curves", rev="9fff22c", features=["derive_serde"] }
+halo2curves = { version = "0.6.0", 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"]}
@@ -34,11 +34,10 @@ 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 = "2.0.11", default_features = false, features = ["ethers-solc"] }
+ethers = { version = "2.0.7", 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" }

src/circuit/ops/base.rs

@@ -125,8 +125,8 @@ impl BaseOp {
             BaseOp::Sum => 1,
             BaseOp::SumInit => 1,
             BaseOp::Range { .. } => 1,
-            BaseOp::IsZero => 0,
-            BaseOp::IsBoolean => 0,
+            BaseOp::IsZero => 1,
+            BaseOp::IsBoolean => 1,
         }
     }
 

src/circuit/ops/chip.rs

@@ -276,20 +276,9 @@ impl<F: PrimeField + TensorType + PartialOrd> BaseConfig<F> {
 
                 let constraints = match base_op {
                     BaseOp::IsBoolean => {
-                        let expected_output: Tensor<Expression<F>> = output
-                            .query_rng(meta, *block_idx, *inner_col_idx, 0, 1)
-                            .expect("non accum: output query failed");
-
-                        let output = expected_output[base_op.constraint_idx()].clone();
-
-                        vec![(output.clone()) * (output.clone() - Expression::Constant(F::from(1)))]
-                    }
-                    BaseOp::IsZero => {
-                        let expected_output: Tensor<Expression<F>> = output
-                            .query_rng(meta, *block_idx, *inner_col_idx, 0, 1)
-                            .expect("non accum: output query failed");
-                        vec![expected_output[base_op.constraint_idx()].clone()]
+                        vec![(qis[1].clone()) * (qis[1].clone() - Expression::Constant(F::from(1)))]
                     }
+                    BaseOp::IsZero => vec![qis[1].clone()],
                     _ => {
                         let expected_output: Tensor<Expression<F>> = output
                             .query_rng(meta, *block_idx, *inner_col_idx, rotation_offset, rng)
@@ -523,10 +512,10 @@ impl<F: PrimeField + TensorType + PartialOrd> BaseConfig<F> {
 
         // we borrow mutably twice so we need to do this dance
 
-        let range_check = if let std::collections::btree_map::Entry::Vacant(e) = self.range_checks.entry(range) {
+        let range_check = if !self.range_checks.contains_key(&range) {
             // as all tables have the same input we see if there's another table who's input we can reuse
             let range_check = RangeCheck::<F>::configure(cs, range);
-            e.insert(range_check.clone());
+            self.range_checks.insert(range, range_check.clone());
             range_check
         } else {
             return Ok(());

src/circuit/ops/hybrid.rs

@@ -6,6 +6,7 @@ use crate::{
     tensor::{self, Tensor, TensorError, TensorType, ValTensor},
 };
 use halo2curves::ff::PrimeField;
+use itertools::Itertools;
 use serde::{Deserialize, Serialize};
 // import run args from model
 
@@ -68,6 +69,14 @@ 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 {
@@ -75,6 +84,7 @@ 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![],
         }
@@ -88,42 +98,176 @@ 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 = 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)
+        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])
+                }
             }
             HybridOp::Recip {
                 input_scale,
                 output_scale,
-                ..
-            } => 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)?,
+                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 indices = Tensor::from(0..x.dims()[*dim] as i128);
+                let mut inter_equals: Vec<Tensor<i128>> = vec![indices.clone(), -indices];
+                let inter =
+                    Op::f(&HybridOp::ReduceMax { axes: vec![*dim] }, inputs)?.intermediate_lookups;
+                inter_equals.extend(inter);
+
+                (res.clone(), inter_equals)
+            }
+            HybridOp::ReduceArgMin { dim } => {
+                let res = tensor::ops::argmin_axes(&x, *dim)?;
+                let indices = Tensor::from(0..x.dims()[*dim] as i128);
+                let mut inter_equals: Vec<Tensor<i128>> = vec![indices.clone(), -indices];
+                let inter =
+                    Op::f(&HybridOp::ReduceMin { axes: vec![*dim] }, inputs)?.intermediate_lookups;
+                inter_equals.extend(inter);
+
+                (res.clone(), inter_equals)
+            }
             HybridOp::Gather { dim, constant_idx } => {
                 if let Some(idx) = constant_idx {
-                    tensor::ops::gather(&x, idx, *dim)?
+                    log::debug!("idx: {}", idx.show());
+                    let res = tensor::ops::gather(&x, idx, *dim)?;
+                    (res.clone(), vec![])
                 } else {
                     let y = inputs[1].clone().map(|x| felt_to_i128(x));
-                    tensor::ops::gather(&x, &y.map(|x| x as usize), *dim)?
+                    let indices = Tensor::from(0..x.dims()[*dim] as i128);
+                    let inter_equals: Vec<Tensor<i128>> = vec![indices.clone(), -indices];
+                    let res = tensor::ops::gather(&x, &y.map(|x| x as usize), *dim)?;
+                    (res.clone(), inter_equals)
                 }
             }
             HybridOp::OneHot { dim, num_classes } => {
-                tensor::ops::one_hot(&x, *num_classes, *dim)?.clone()
+                let indices = Tensor::from(0..x.dims()[*dim] as i128);
+                let inter_equals: Vec<Tensor<i128>> = vec![indices.clone(), -indices];
+                let res = tensor::ops::one_hot(&x, *num_classes, *dim)?;
+                (res.clone(), inter_equals)
             }
+            HybridOp::TopK { dim, k, largest } => {
+                let res = tensor::ops::topk_axes(&x, *k, *dim, *largest)?;
 
-            HybridOp::TopK { dim, k, largest } => 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 indices = Tensor::from(0..x.dims()[*dim] as i128);
+                    let inter_equals: Vec<Tensor<i128>> = vec![indices.clone(), -indices];
+                    let res = tensor::ops::gather_elements(&x, &y.map(|x| x as usize), *dim)?;
+                    (res.clone(), inter_equals)
+                }
+            }
+            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 indices = Tensor::from(0..x.dims()[*dim] as i128);
+                    let inter_equals: Vec<Tensor<i128>> = vec![indices.clone(), -indices];
+                    let res = tensor::ops::scatter(&x, &idx, &src, *dim)?;
+                    (res.clone(), inter_equals)
+                }
+            }
             HybridOp::MaxPool2d {
                 padding,
                 stride,
                 pool_dims,
                 ..
-            } => tensor::ops::max_pool2d(&x, 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],
+                )
+            }
             HybridOp::SumPool {
                 padding,
                 stride,
@@ -135,7 +279,10 @@ 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)
+                (
+                    tensor::ops::nonlinearities::range_check_percent(&[x, y], 128, 128, tol.val),
+                    vec![],
+                )
             }
             HybridOp::Greater => {
                 let y = inputs[1].clone().map(|x| felt_to_i128(x));
@@ -162,7 +309,10 @@ 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 })
+        Ok(ForwardResult {
+            output,
+            intermediate_lookups,
+        })
     }
 
     fn as_string(&self) -> String {
@@ -216,6 +366,8 @@ 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)
             }
@@ -288,7 +440,9 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for HybridOp {
                         config,
                         region,
                         values.try_into()?,
-                        &LookupOp::Div { denom: *denom },
+                        &LookupOp::Div {
+                            denom: denom.clone(),
+                        },
                     )?
                 }
             }
@@ -299,7 +453,26 @@ 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

@@ -128,7 +128,13 @@ pub fn recip<F: PrimeField + TensorType + PartialOrd>(
 
     let range_check_bracket = felt_to_i128(output_scale * input_scale) / 2;
 
-    let is_assigned = !input.any_unknowns()?;
+    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 mut claimed_output: ValTensor<F> = if is_assigned {
         let input_evals = input.get_int_evals()?;
@@ -160,14 +166,27 @@ 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,
-        &[product],
-        &(range_check_bracket, 3 * range_check_bracket),
+        &[diff_with_input],
+        &(-range_check_bracket, range_check_bracket),
     )?;
 
     Ok(claimed_output)
@@ -214,7 +233,7 @@ 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, ValType::Constant(F::ZERO))?;
+        input.pad_to_zero_rem(block_width)?;
         let inp = {
             let (res, len) = region.assign_with_duplication(
                 &config.inputs[i],
@@ -770,7 +789,14 @@ fn one_hot<F: PrimeField + TensorType + PartialOrd>(
     let assigned_input = region.assign(&config.inputs[0], &input)?;
 
     // now assert all elems are 0 or 1
-    let assigned_output = boolean_identity(config, region, &[output.clone()], true)?;
+    let assigned_output = region.assign(&config.inputs[1], &output)?;
+    if !region.is_dummy() {
+        for i in 0..assigned_output.len() {
+            let (x, y, z) = config.output.cartesian_coord(region.linear_coord() + i);
+            let selector = config.selectors.get(&(BaseOp::IsBoolean, x, y));
+            region.enable(selector, z)?;
+        }
+    }
     region.increment(std::cmp::max(assigned_output.len(), assigned_input.len()));
 
     let sum = sum(config, region, &[assigned_output.clone()])?;
@@ -1142,7 +1168,7 @@ 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, ValType::Constant(F::ZERO))?;
+        input.pad_to_zero_rem(block_width)?;
         let (res, len) =
             region.assign_with_duplication(&config.inputs[1], &input, &config.check_mode, false)?;
         assigned_len = len;
@@ -1211,7 +1237,7 @@ 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, ValType::Constant(F::ONE))?;
+        input.pad_to_zero_rem(block_width)?;
         let (res, len) =
             region.assign_with_duplication(&config.inputs[1], &input, &config.check_mode, false)?;
         assigned_len = len;
@@ -1675,28 +1701,10 @@ pub fn equals<F: PrimeField + TensorType + PartialOrd>(
     values: &[ValTensor<F>; 2],
 ) -> Result<ValTensor<F>, Box<dyn Error>> {
     let diff = pairwise(config, region, values, BaseOp::Sub)?;
-    let diff_inverse = diff.inverse()?;
-    let product_diff_and_invert =
-        pairwise(config, region, &[diff.clone(), diff_inverse], BaseOp::Mult)?;
 
-    // constant of 1
-    let mut ones = Tensor::from(vec![ValType::Constant(F::from(1))].into_iter());
-    ones.set_visibility(&crate::graph::Visibility::Fixed);
+    let res = nonlinearity(config, region, &[diff], &LookupOp::KroneckerDelta)?;
 
-    // subtract
-    let output = pairwise(
-        config,
-        region,
-        &[ones.into(), product_diff_and_invert],
-        BaseOp::Sub,
-    )?;
-
-    // take the product of diff and output
-    let prod_check = pairwise(config, region, &[diff, output.clone()], BaseOp::Mult)?;
-
-    is_zero_identity(config, region, &[prod_check], false)?;
-
-    Ok(output)
+    Ok(res)
 }
 
 /// Xor boolean operation
@@ -1760,7 +1768,21 @@ pub fn iff<F: PrimeField + TensorType + PartialOrd>(
             .into();
 
     // make sure mask is boolean
-    let assigned_mask = boolean_identity(config, region, &[mask.clone()], true)?;
+    let assigned_mask = region.assign(&config.inputs[1], mask)?;
+
+    // Enable the selectors
+    if !region.is_dummy() {
+        (0..assigned_mask.len())
+            .map(|i| {
+                let (x, y, z) = config.inputs[1].cartesian_coord(region.linear_coord() + i);
+                let selector = config.selectors.get(&(BaseOp::IsBoolean, x, y));
+                region.enable(selector, z)?;
+                Ok(())
+            })
+            .collect::<Result<Vec<_>, Box<dyn Error>>>()?;
+    }
+
+    region.increment(assigned_mask.len());
 
     let one_minus_mask = pairwise(config, region, &[unit, assigned_mask.clone()], BaseOp::Sub)?;
 
@@ -1858,11 +1880,13 @@ pub fn sumpool<F: PrimeField + TensorType + PartialOrd>(
     last_elem.reshape(&[&[batch_size, image_channels], shape].concat())?;
 
     if normalized {
-        last_elem = div(
+        last_elem = nonlinearity(
             config,
             region,
             &[last_elem],
-            F::from((kernel_shape.0 * kernel_shape.1) as u64),
+            &LookupOp::Div {
+                denom: utils::F32((kernel_shape.0 * kernel_shape.1) as f32),
+            },
         )?;
     }
     Ok(last_elem)
@@ -2359,60 +2383,18 @@ pub fn identity<F: PrimeField + TensorType + PartialOrd>(
     Ok(output)
 }
 
-/// is zero identity constraint. Usually used to constrain an instance column to an advice so the returned cells / values can be operated upon.
-pub fn is_zero_identity<F: PrimeField + TensorType + PartialOrd>(
-    config: &BaseConfig<F>,
-    region: &mut RegionCtx<F>,
-    values: &[ValTensor<F>; 1],
-    assign: bool,
-) -> Result<ValTensor<F>, Box<dyn Error>> {
-    let output = if assign || !values[0].get_const_indices()?.is_empty() {
-        let output = region.assign(&config.output, &values[0])?;
-        region.increment(output.len());
-        output
-    } else {
-        values[0].clone()
-    };
-    // Enable the selectors
-    if !region.is_dummy() {
-        (0..output.len())
-            .map(|j| {
-                let index = region.linear_coord() - j - 1;
-
-                let (x, y, z) = config.output.cartesian_coord(index);
-                let selector = config.selectors.get(&(BaseOp::IsZero, x, y));
-
-                region.enable(selector, z)?;
-                Ok(())
-            })
-            .collect::<Result<Vec<_>, Box<dyn Error>>>()?;
-    }
-
-    Ok(output)
-}
-
 /// Boolean identity constraint. Usually used to constrain an instance column to an advice so the returned cells / values can be operated upon.
 pub fn boolean_identity<F: PrimeField + TensorType + PartialOrd>(
     config: &BaseConfig<F>,
     region: &mut RegionCtx<F>,
     values: &[ValTensor<F>; 1],
-    assign: bool,
 ) -> Result<ValTensor<F>, Box<dyn Error>> {
-    let output = if assign || !values[0].get_const_indices()?.is_empty() {
-        // get zero constants indices
-        let output = region.assign(&config.output, &values[0])?;
-        region.increment(output.len());
-        output
-    } else {
-        values[0].clone()
-    };
+    let output = region.assign(&config.inputs[1], &values[0])?;
     // Enable the selectors
     if !region.is_dummy() {
         (0..output.len())
             .map(|j| {
-                let index = region.linear_coord() - j - 1;
-
-                let (x, y, z) = config.output.cartesian_coord(index);
+                let (x, y, z) = config.inputs[1].cartesian_coord(region.linear_coord() + j);
                 let selector = config.selectors.get(&(BaseOp::IsBoolean, x, y));
 
                 region.enable(selector, z)?;
@@ -2420,6 +2402,7 @@ pub fn boolean_identity<F: PrimeField + TensorType + PartialOrd>(
             })
             .collect::<Result<Vec<_>, Box<dyn Error>>>()?;
     }
+    region.increment(output.len());
 
     Ok(output)
 }
@@ -2469,7 +2452,7 @@ 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();
@@ -2488,7 +2471,7 @@ pub fn range_check<F: PrimeField + TensorType + PartialOrd>(
                 let (x, y, z) = config
                     .lookup_input
                     .cartesian_coord(region.linear_coord() + i);
-                let selector = config.range_check_selectors.get(&(*range, x, y));
+                let selector = config.range_check_selectors.get(&(range.clone(), x, y));
                 region.enable(selector, z)?;
                 Ok(())
             })
@@ -2515,7 +2498,7 @@ 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(), values)?;
+    region.add_used_lookup(nl.clone());
 
     // time the entire operation
     let timer = instant::Instant::now();
@@ -2598,6 +2581,22 @@ pub fn nonlinearity<F: PrimeField + TensorType + PartialOrd>(
     Ok(output)
 }
 
+/// mean function layout
+pub fn mean<F: PrimeField + TensorType + PartialOrd>(
+    config: &BaseConfig<F>,
+    region: &mut RegionCtx<F>,
+    values: &[ValTensor<F>; 1],
+    scale: usize,
+) -> Result<ValTensor<F>, Box<dyn Error>> {
+    let x = &values[0];
+
+    let sum_x = sum(config, region, &[x.clone()])?;
+    let nl = LookupOp::Div {
+        denom: utils::F32((scale * x.len()) as f32),
+    };
+    nonlinearity(config, region, &[sum_x], &nl)
+}
+
 /// Argmax
 pub fn argmax<F: PrimeField + TensorType + PartialOrd>(
     config: &BaseConfig<F>,
@@ -2710,8 +2709,24 @@ pub fn max<F: PrimeField + TensorType + PartialOrd>(
     )?;
     // relu(x - max(x - 1))
     let relu = nonlinearity(config, region, &[diff], &LookupOp::ReLU)?;
-    // constraining relu(x - max(x - 1)) = 0/1
-    boolean_identity(config, region, &[relu.clone()], false)?;
+
+    let len = relu.dims().iter().product();
+
+    // y_i*(1 - y_i) =0 // assert the values are either 0 or 1
+    region.assign(&config.inputs[1], &relu)?;
+
+    if !region.is_dummy() {
+        (0..len)
+            .map(|i| {
+                let (x, y, z) = config.inputs[1].cartesian_coord(region.linear_coord() + i);
+                let selector = config.selectors.get(&(BaseOp::IsBoolean, x, y));
+                region.enable(selector, z)?;
+                Ok(())
+            })
+            .collect::<Result<Vec<_>, Box<dyn Error>>>()?;
+    }
+
+    region.increment(len);
 
     // sum(relu(x - max(x - 1)))
     let sum_relu = sum(config, region, &[relu])?;
@@ -2722,7 +2737,13 @@ pub fn max<F: PrimeField + TensorType + PartialOrd>(
         nonlinearity(config, region, &[one_minus_sum_relu], &LookupOp::ReLU)?;
 
     // constraining 1 - sum(relu(x - max(x - 1))) = 0
-    is_zero_identity(config, region, &[relu_one_minus_sum_relu], false)?;
+    region.assign(&config.inputs[1], &relu_one_minus_sum_relu)?;
+
+    let (x, y, z) = config.output.cartesian_coord(region.linear_coord());
+    let selector = config.selectors.get(&(BaseOp::IsZero, x, y));
+    region.enable(selector, z)?;
+
+    region.increment(relu_one_minus_sum_relu.len());
 
     Ok(assigned_max_val)
 }
@@ -2767,8 +2788,23 @@ pub fn min<F: PrimeField + TensorType + PartialOrd>(
 
     // relu(min(x + 1)  - x)
     let relu = nonlinearity(config, region, &[diff], &LookupOp::ReLU)?;
-    // constraining relu(min(x + 1) - x) = 0/1
-    boolean_identity(config, region, &[relu.clone()], false)?;
+
+    let len = relu.dims().iter().product();
+
+    region.assign(&config.inputs[1], &relu)?;
+    // y_i*(1 - y_i) =0 // assert the values are either 0 or 1
+    if !region.is_dummy() {
+        (0..len)
+            .map(|i| {
+                let (x, y, z) = config.inputs[1].cartesian_coord(region.linear_coord() + i);
+                let selector = config.selectors.get(&(BaseOp::IsBoolean, x, y));
+                region.enable(selector, z)?;
+                Ok(())
+            })
+            .collect::<Result<Vec<_>, Box<dyn Error>>>()?;
+    }
+
+    region.increment(len);
 
     // sum(relu(min(x + 1) - x))
     let sum_relu = sum(config, region, &[relu])?;
@@ -2779,8 +2815,14 @@ pub fn min<F: PrimeField + TensorType + PartialOrd>(
     let relu_one_minus_sum_relu =
         nonlinearity(config, region, &[one_minus_sum_relu], &LookupOp::ReLU)?;
 
+    region.assign(&config.inputs[1], &relu_one_minus_sum_relu)?;
+
     // constraining product to 0
-    is_zero_identity(config, region, &[relu_one_minus_sum_relu], false)?;
+    let (x, y, z) = config.output.cartesian_coord(region.linear_coord());
+    let selector = config.selectors.get(&(BaseOp::IsZero, x, y));
+    region.enable(selector, z)?;
+
+    region.increment(relu_one_minus_sum_relu.len());
 
     Ok(assigned_min_val)
 }
@@ -2999,8 +3041,15 @@ pub fn range_check_percent<F: PrimeField + TensorType + PartialOrd>(
     // Add the lower_bound and upper_bound
     let sum = pairwise(config, region, &[lower_bound, upper_bound], BaseOp::Add)?;
 
+    // Assign the sum tensor to the inputs
+    region.assign(&config.inputs[1], &sum)?;
+
     // Constrain the sum to be all zeros
-    is_zero_identity(config, region, &[sum.clone()], false)?;
+    let (x, y, z) = config.output.cartesian_coord(region.linear_coord());
+    let selector = config.selectors.get(&(BaseOp::IsZero, x, y));
+    region.enable(selector, z)?;
+
+    region.increment(sum.len());
 
     Ok(sum)
 }

src/circuit/ops/lookup.rs

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

src/circuit/ops/mod.rs

@@ -29,6 +29,7 @@ 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.
@@ -177,6 +178,7 @@ 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![],
         })
     }
 
@@ -199,7 +201,6 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for Input {
                         config,
                         region,
                         values[..].try_into()?,
-                        true,
                     )?))
                 }
                 _ => Ok(Some(super::layouts::identity(
@@ -302,7 +303,10 @@ 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 })
+        Ok(ForwardResult {
+            output,
+            intermediate_lookups: vec![],
+        })
     }
 
     fn as_string(&self) -> String {

src/circuit/ops/poly.rs

@@ -1,6 +1,5 @@
 use crate::{
     circuit::layouts,
-    fieldutils::felt_to_i128,
     tensor::{self, Tensor, TensorError},
 };
 
@@ -10,14 +9,6 @@ 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>,
     },
@@ -90,8 +81,6 @@ 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(),
@@ -214,36 +203,14 @@ impl<F: PrimeField + TensorType + PartialOrd + Serialize + for<'de> Deserialize<
                 if 1 != inputs.len() {
                     return Err(TensorError::DimMismatch("slice inputs".to_string()));
                 }
-                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(tensor::ops::slice(&inputs[0], axis, start, end)?)
             }
         }?;
 
-        Ok(ForwardResult { output: res })
+        Ok(ForwardResult {
+            output: res,
+            intermediate_lookups: vec![],
+        })
     }
 
     fn layout(
@@ -284,26 +251,6 @@ 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,
@@ -405,8 +352,6 @@ 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

@@ -16,8 +16,6 @@ use std::{
     },
 };
 
-use portable_atomic::AtomicI128 as AtomicInt;
-
 use super::lookup::LookupOp;
 
 /// Region error
@@ -68,8 +66,6 @@ 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> {
@@ -91,8 +87,6 @@ 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
@@ -110,8 +104,6 @@ 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,
         }
     }
 
@@ -128,8 +120,6 @@ 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,
         }
     }
 
@@ -151,8 +141,6 @@ impl<'a, F: PrimeField + TensorType + PartialOrd> RegionCtx<'a, F> {
             total_constants,
             used_lookups,
             used_range_checks,
-            max_lookup_inputs: 0,
-            min_lookup_inputs: 0,
         }
     }
 
@@ -192,7 +180,6 @@ 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>,
@@ -203,10 +190,6 @@ 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()));
 
@@ -238,14 +221,6 @@ 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());
@@ -259,11 +234,6 @@ 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)))?
@@ -283,54 +253,19 @@ 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,
-        inputs: &[ValTensor<F>],
-    ) -> Result<(), Box<dyn std::error::Error>> {
+    pub fn add_used_lookup(&mut self, lookup: LookupOp) {
         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) -> Result<(), Box<dyn std::error::Error>> {
+    pub fn add_used_range_check(&mut self, range: Range) {
         self.used_range_checks.insert(range);
-        self.update_max_min_lookup_range(range)
     }
 
     /// Get the offset
@@ -358,16 +293,6 @@ 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/execute.rs

@@ -726,11 +726,11 @@ impl AccuracyResults {
             let percentage_error = error.enum_map(|i, x| Ok::<_, TensorError>(x / original[i]))?;
             let abs_percentage_error = percentage_error.map(|x| x.abs());
 
-            errors.extend(error);
-            abs_errors.extend(abs_error);
-            squared_errors.extend(squared_error);
-            percentage_errors.extend(percentage_error);
-            abs_percentage_errors.extend(abs_percentage_error);
+            errors.extend(error.into_iter());
+            abs_errors.extend(abs_error.into_iter());
+            squared_errors.extend(squared_error.into_iter());
+            percentage_errors.extend(percentage_error.into_iter());
+            abs_percentage_errors.extend(abs_percentage_error.into_iter());
         }
 
         let mean_percent_error =
@@ -826,7 +826,10 @@ pub(crate) fn calibrate(
     let range = if let Some(scales) = scales {
         scales
     } else {
-        (10..14).collect::<Vec<crate::Scale>>()
+        match target {
+            CalibrationTarget::Resources { .. } => (8..10).collect::<Vec<crate::Scale>>(),
+            CalibrationTarget::Accuracy => (10..14).collect::<Vec<crate::Scale>>(),
+        }
     };
 
     let div_rebasing = if let Some(div_rebasing) = div_rebasing {

src/graph/model.rs

@@ -10,6 +10,7 @@ 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},
@@ -67,16 +68,6 @@ 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 {
@@ -102,12 +93,6 @@ 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.
@@ -501,25 +486,7 @@ impl Model {
         );
         // this is the total number of variables we will need to allocate
         // for the circuit
-        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)?;
+        let res = self.dummy_layout(run_args, &self.graph.input_shapes()?)?;
 
         // if we're using percentage tolerance, we need to add the necessary range check ops for it.
 
@@ -555,12 +522,206 @@ 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 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())
+        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)
     }
 
     /// Loads an Onnx model from a specified path.
@@ -1012,8 +1173,8 @@ impl Model {
         );
         // set scale for HybridOp::RangeCheck and call self.conf_lookup on that op for percentage tolerance case
         let input = &vars.advices[0];
-        let output = &vars.advices[2];
-        let index = &vars.advices[1];
+        let output = &vars.advices[1];
+        let index = &vars.advices[2];
         for op in required_lookups {
             base_gate.configure_lookup(meta, input, output, index, lookup_range, logrows, &op)?;
         }
@@ -1174,29 +1335,18 @@ impl Model {
             };
 
             debug!(
-                "laying out {}: {}, row:{}, coord:{}, total_constants: {}, max_lookup_inputs: {}, min_lookup_inputs: {}",
+                "laying out {}: {}, row:{}, coord:{}, total_constants: {}",
                 idx,
                 node.as_str(),
                 region.row(),
                 region.linear_coord(),
-                region.total_constants(),
-                region.max_lookup_inputs(),
-                region.min_lookup_inputs()
+                region.total_constants()
             );
             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) => {
@@ -1339,13 +1489,28 @@ impl Model {
     pub fn dummy_layout(
         &self,
         run_args: &RunArgs,
-        inputs: &[ValTensor<Fp>],
+        input_shapes: &[Vec<usize>],
     ) -> Result<DummyPassRes, Box<dyn Error>> {
         info!("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()]);
@@ -1380,12 +1545,12 @@ impl Model {
                 .collect::<Result<Vec<_>, Box<dyn Error>>>()?;
 
             let _ = outputs
-                .iter()
+                .into_iter()
                 .zip(comparator)
                 .map(|(o, c)| {
                     dummy_config.layout(
                         &mut region,
-                        &[o.clone(), c],
+                        &[o, c],
                         Box::new(HybridOp::RangeCheck(run_args.tolerance)),
                     )
                 })
@@ -1410,23 +1575,12 @@ 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

@@ -148,7 +148,7 @@ impl RebaseScale {
                 SupportedOp::RebaseScale(RebaseScale {
                     inner: op.inner.clone(),
                     target_scale: op.target_scale,
-                    multiplier,
+                    multiplier: multiplier,
                     rebase_op: HybridOp::Div {
                         denom: crate::circuit::utils::F32((multiplier) as f32),
                         use_range_check_for_int: !div_rebasing,
@@ -219,6 +219,8 @@ 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

@@ -439,16 +439,17 @@ 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::Linear(crate::circuit::ops::poly::PolyOp::ScatterElements {
-                dim: axis,
-                constant_idx: None,
-            });
+            let mut op =
+                SupportedOp::Hybrid(crate::circuit::ops::hybrid::HybridOp::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::Linear(crate::circuit::ops::poly::PolyOp::ScatterElements {
+                op = SupportedOp::Hybrid(crate::circuit::ops::hybrid::HybridOp::ScatterElements {
                     dim: axis,
                     constant_idx: Some(c.raw_values.map(|x| x as usize)),
                 })
@@ -477,16 +478,17 @@ 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::Linear(crate::circuit::ops::poly::PolyOp::GatherElements {
-                dim: axis,
-                constant_idx: None,
-            });
+            let mut op =
+                SupportedOp::Hybrid(crate::circuit::ops::hybrid::HybridOp::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::Linear(crate::circuit::ops::poly::PolyOp::GatherElements {
+                op = SupportedOp::Hybrid(crate::circuit::ops::hybrid::HybridOp::GatherElements {
                     dim: axis,
                     constant_idx: Some(c.raw_values.map(|x| x as usize)),
                 })

src/lib.rs

@@ -7,6 +7,7 @@
     overflowing_literals,
     path_statements,
     patterns_in_fns_without_body,
+    private_in_public,
     unconditional_recursion,
     unused,
     unused_allocation,

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, 0).unwrap(), expected);
+    /// assert_eq!(a.pad_to_zero_rem(4).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, 0).unwrap(), expected);
+    /// assert_eq!(a.pad_to_zero_rem(9).unwrap(), expected);
     /// ```
-    pub fn pad_to_zero_rem(&self, n: usize, pad: T) -> Result<Tensor<T>, TensorError> {
+    pub fn pad_to_zero_rem(&self, n: usize) -> Result<Tensor<T>, TensorError> {
         let mut inner = self.inner.clone();
         let remainder = self.len() % n;
         if remainder != 0 {
-            inner.resize(self.len() + n - remainder, pad);
+            inner.resize(self.len() + n - remainder, T::zero().unwrap());
         }
         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();
+/// let result = equals(&a, &b).unwrap().0;
 /// 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>, TensorError> {
+) -> Result<(Tensor<T>, Vec<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)
+    Ok((result, vec![diff]))
 }
 
 /// 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, expected);
+/// assert_eq!(result.0, expected);
 /// ```
 pub fn greater<
     T: TensorType
@@ -302,7 +302,7 @@ pub fn greater<
 >(
     a: &Tensor<T>,
     b: &Tensor<T>,
-) -> Result<Tensor<T>, TensorError> {
+) -> Result<(Tensor<T>, Vec<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)
+    Ok((mask, vec![mask_inter]))
 }
 
 /// 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, expected);
+/// assert_eq!(result.0, expected);
 /// ```
 pub fn greater_equal<
     T: TensorType
@@ -345,7 +345,7 @@ pub fn greater_equal<
 >(
     a: &Tensor<T>,
     b: &Tensor<T>,
-) -> Result<Tensor<T>, TensorError> {
+) -> Result<(Tensor<T>, Vec<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)
+    Ok((mask, vec![mask_inter]))
 }
 
 /// 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, expected);
+/// assert_eq!(result.0, expected);
 /// ```
 ///
 pub fn less<
@@ -389,7 +389,7 @@ pub fn less<
 >(
     a: &Tensor<T>,
     b: &Tensor<T>,
-) -> Result<Tensor<T>, TensorError> {
+) -> Result<(Tensor<T>, Vec<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, expected);
+/// assert_eq!(result.0, expected);
 /// ```
 ///
 pub fn less_equal<
@@ -426,7 +426,7 @@ pub fn less_equal<
 >(
     a: &Tensor<T>,
     b: &Tensor<T>,
-) -> Result<Tensor<T>, TensorError> {
+) -> Result<(Tensor<T>, Vec<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();
+/// let pooled = sumpool(&x, [(0, 0); 2], (1, 1), (2, 2), false).unwrap().0;
 /// 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();
+/// let pooled = sumpool(&x, [(0, 0); 2], (1, 1), (2, 2), true).unwrap().0;
 /// 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>, TensorError> {
+) -> Result<(Tensor<i128>, Vec<Tensor<i128>>), TensorError> {
     let image_dims = image.dims();
     let batch_size = image_dims[0];
     let image_channels = image_dims[1];
@@ -2345,12 +2345,15 @@ 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)
+    Ok((combined, inter))
 }
 
 /// Applies 2D max pooling over a 4D tensor of shape B x C x H x W.
@@ -3045,7 +3048,11 @@ pub mod nonlinearities {
     }
 
     /// softmax layout
-    pub fn softmax_axes(a: &Tensor<i128>, scale: f64, axes: &[usize]) -> Tensor<i128> {
+    pub fn softmax_axes(
+        a: &Tensor<i128>,
+        scale: f64,
+        axes: &[usize],
+    ) -> (Tensor<i128>, Vec<Tensor<i128>>) {
         // we want this to be as small as possible so we set the output scale to 1
         let dims = a.dims();
 
@@ -3053,6 +3060,8 @@ pub mod nonlinearities {
             return softmax(a, scale);
         }
 
+        let mut intermediate_values = vec![];
+
         let cartesian_coord = dims[..dims.len() - 1]
             .iter()
             .map(|x| 0..*x)
@@ -3075,7 +3084,8 @@ pub mod nonlinearities {
 
             let res = softmax(&softmax_input, scale);
 
-            outputs.push(res);
+            outputs.push(res.0);
+            intermediate_values.extend(res.1);
         }
 
         let mut res = Tensor::new(Some(&outputs), &[outputs.len()])
@@ -3083,7 +3093,7 @@ pub mod nonlinearities {
             .combine()
             .unwrap();
         res.reshape(dims).unwrap();
-        res
+        (res, intermediate_values)
     }
 
     /// Applies softmax
@@ -3100,20 +3110,24 @@ pub mod nonlinearities {
     ///     Some(&[2, 2, 3, 2, 2, 0]),
     ///     &[2, 3],
     /// ).unwrap();
-    /// let result = softmax(&x, 128.0);
+    /// let result = softmax(&x, 128.0).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> {
+    pub fn softmax(a: &Tensor<i128>, scale: f64) -> (Tensor<i128>, Vec<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()
+        ((exp * inv_denom).unwrap(), intermediate_values)
     }
 
     /// 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, pad: ValType<F>) -> Result<(), Box<dyn Error>> {
+    pub fn pad_to_zero_rem(&mut self, n: usize) -> Result<(), Box<dyn Error>> {
         match self {
             ValTensor::Value {
                 inner: v, dims: d, ..
             } => {
-                *v = v.pad_to_zero_rem(n, pad)?;
+                *v = v.pad_to_zero_rem(n)?;
                 *d = v.dims().to_vec();
             }
             ValTensor::Instance { .. } => {
@@ -871,30 +871,3 @@ impl<F: PrimeField + TensorType + PartialOrd> ValTensor<F> {
         }
     }
 }
-
-impl<F: PrimeField + TensorType + PartialOrd> ValTensor<F> {
-    /// inverts the inner values
-    pub fn inverse(&self) -> Result<ValTensor<F>, Box<dyn Error>> {
-        let mut cloned_self = self.clone();
-
-        match &mut cloned_self {
-            ValTensor::Value {
-                inner: v, dims: d, ..
-            } => {
-                *v = v.map(|x| match x {
-                    ValType::AssignedValue(v) => ValType::AssignedValue(v.invert()),
-                    ValType::PrevAssigned(v) | ValType::AssignedConstant(v, ..) => {
-                        ValType::AssignedValue(v.value_field().invert())
-                    }
-                    ValType::Value(v) => ValType::Value(v.map(|x| x.invert().unwrap_or(F::ZERO))),
-                    ValType::Constant(v) => ValType::Constant(v.invert().unwrap_or(F::ZERO)),
-                });
-                *d = v.dims().to_vec();
-            }
-            ValTensor::Instance { .. } => {
-                return Err(Box::new(TensorError::WrongMethod));
-            }
-        };
-        Ok(cloned_self)
-    }
-}

tests/integration_tests.rs

@@ -192,86 +192,86 @@ mod native_tests {
     ];
 
     const TESTS: [&str; 77] = [
-        "1l_mlp", //0
+        "1l_mlp",
         "1l_slice",
         "1l_concat",
         "1l_flatten",
         // "1l_average",
         "1l_div",
-        "1l_pad", // 5
+        "1l_pad",
         "1l_reshape",
         "1l_eltwise_div",
         "1l_sigmoid",
         "1l_sqrt",
-        "1l_softmax", //10
+        "1l_softmax",
         // "1l_instance_norm",
         "1l_batch_norm",
         "1l_prelu",
         "1l_leakyrelu",
         "1l_gelu_noappx",
         // "1l_gelu_tanh_appx",
-        "1l_relu", //15
+        "1l_relu",
         "1l_downsample",
         "1l_tanh",
         "2l_relu_sigmoid_small",
         "2l_relu_fc",
-        "2l_relu_small", //20
+        "2l_relu_small",
         "2l_relu_sigmoid",
         "1l_conv",
         "2l_sigmoid_small",
         "2l_relu_sigmoid_conv",
-        "3l_relu_conv_fc", //25
+        "3l_relu_conv_fc",
         "4l_relu_conv_fc",
         "1l_erf",
         "1l_var",
-        "1l_elu",
-        "min", //30
+        "1l_elu", //30
+        "min",
         "max",
         "1l_max_pool",
         "1l_conv_transpose",
-        "1l_upsample",
-        "1l_identity", //35
+        "1l_upsample", //35
+        "1l_identity",
         "idolmodel",
         "trig",
         "prelu_gmm",
-        "lstm",
-        "rnn", //40
+        "lstm", //40
+        "rnn",
         "quantize_dequantize",
         "1l_where",
         "boolean",
         "boolean_identity",
-        "decision_tree", // 45
+        "decision_tree", // "variable_cnn",
         "random_forest",
         "gradient_boosted_trees",
         "1l_topk",
-        "xgboost",
-        "lightgbm", //50
+        "xgboost", //50
+        "lightgbm",
         "hummingbird_decision_tree",
         "oh_decision_tree",
         "linear_svc",
         "gather_elements",
-        "less", //55
+        "less",
         "xgboost_reg",
         "1l_powf",
         "scatter_elements",
-        "1l_linear",
-        "linear_regression", //60
+        "1l_linear", //60
+        "linear_regression",
         "sklearn_mlp",
         "1l_mean",
         "rounding_ops",
         // "mean_as_constrain",
         "arange",
-        "layernorm", //65
+        "layernorm",
         "bitwise_ops",
         "blackman_window",
-        "softsign", //68
+        "softsign", //70
         "softplus",
-        "selu", //70
+        "selu",
         "hard_sigmoid",
         "log_softmax",
         "eye",
         "ltsf",
-        "remainder", //75
+        "remainder",
         "bitshift",
     ];
 
@@ -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", 3.1, false);
+                accuracy_measurement(path, test.to_string(), "private", "private", "public", 1, "resources", 18.0, false);
                 test_dir.close().unwrap();
             }
 

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", // 12
+        "set_membership.ipynb",
         "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_relu',
+        '1l_average',
         'input.json'
     )
     model_path = os.path.join(
         examples_path,
         'onnx',
-        '1l_relu',
+        '1l_average',
         'network.onnx'
     )
     output_path = os.path.join(
@@ -147,13 +147,13 @@ def test_calibrate():
     data_path = os.path.join(
         examples_path,
         'onnx',
-        '1l_relu',
+        '1l_average',
         'input.json'
     )
     model_path = os.path.join(
         examples_path,
         'onnx',
-        '1l_relu',
+        '1l_average',
         'network.onnx'
     )
     output_path = os.path.join(
@@ -183,7 +183,7 @@ def test_model_compile():
     model_path = os.path.join(
         examples_path,
         'onnx',
-        '1l_relu',
+        '1l_average',
         'network.onnx'
     )
     compiled_model_path = os.path.join(
@@ -205,7 +205,7 @@ def test_forward():
     data_path = os.path.join(
         examples_path,
         'onnx',
-        '1l_relu',
+        '1l_average',
         'input.json'
     )
     model_path = os.path.join(