feat: lookup-less recip by default (#725)

BREAKING CHANGE: python and wasm felt utilities have new names

.github/workflows/large-tests.yml

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

.github/workflows/release.yml

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

.github/workflows/rust.yml

@@ -26,7 +26,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - name: Build
@@ -38,7 +38,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - name: Docs
@@ -50,7 +50,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -73,7 +73,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -106,7 +106,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -139,7 +139,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -172,7 +172,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -189,7 +189,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: jetli/wasm-pack-action@v0.4.0
@@ -198,10 +198,8 @@ 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-2023-08-24-x86_64-unknown-linux-gnu
+        run: rustup component add rust-src --toolchain nightly-2024-01-04-x86_64-unknown-linux-gnu
       - name: Run wasm verifier tests
         # on mac:
         #   AR=/opt/homebrew/opt/llvm/bin/llvm-ar CC=/opt/homebrew/opt/llvm/bin/clang wasm-pack test --firefox --headless -- -Z build-std="panic_abort,std" --features web
@@ -214,7 +212,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -231,13 +229,15 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
         with:
           crate: cargo-nextest
           locked: true
+      - name: public outputs and tolerance > 0
+        run: cargo nextest run --release --verbose tests::mock_tolerance_public_outputs_ --test-threads 32
       - name: public outputs + batch size == 10
         run: cargo nextest run --release --verbose tests::mock_large_batch_public_outputs_ --test-threads 32
       - name: kzg inputs
@@ -286,7 +286,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -312,7 +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 95a93cd397f25f3f8d49d2851eb52bc2d52dd983 --profile local --locked anvil --force
+        run: cargo install --git https://github.com/foundry-rs/foundry --rev b320f350156a0fb15c2eb13dc380deb2367c4474 --profile local --locked anvil --force
       - name: KZG prove and verify tests (EVM + VK rendered seperately)
         run: cargo nextest run --release --verbose tests_evm::kzg_evm_prove_and_verify_render_seperately_ --test-threads 1
       - name: KZG prove and verify tests (EVM + kzg all)
@@ -345,18 +345,15 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: jetli/wasm-pack-action@v0.4.0
       - 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-2023-08-24-x86_64-unknown-linux-gnu
+        run: rustup component add rust-src --toolchain nightly-2024-01-04-x86_64-unknown-linux-gnu
       - uses: actions/checkout@v3
       - name: Use pnpm 8
         uses: pnpm/action-setup@v2
@@ -416,11 +413,11 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - name: Add rust-src
-        run: rustup component add rust-src --toolchain nightly-2023-08-24-x86_64-unknown-linux-gnu
+        run: rustup component add rust-src --toolchain nightly-2024-01-04-x86_64-unknown-linux-gnu
       - uses: actions/checkout@v3
       - uses: baptiste0928/cargo-install@v1
         with:
@@ -450,7 +447,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -460,7 +457,7 @@ jobs:
       - name: Install solc
         run: (hash svm 2>/dev/null || cargo install svm-rs) && svm install 0.8.20 && solc --version
       - name: Install Anvil
-        run: cargo install --git https://github.com/foundry-rs/foundry --rev 95a93cd397f25f3f8d49d2851eb52bc2d52dd983 --profile local --locked anvil --force
+        run: cargo install --git https://github.com/foundry-rs/foundry --rev b320f350156a0fb15c2eb13dc380deb2367c4474 --profile local --locked anvil --force
       - name: fuzz tests (EVM)
         run: cargo nextest run --release --verbose tests_evm::kzg_evm_fuzz_ --test-threads 2
       # - name: fuzz tests
@@ -473,7 +470,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -491,7 +488,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -503,12 +500,12 @@ jobs:
 
   prove-and-verify-aggr-tests:
     runs-on: large-self-hosted
-    needs: [build, library-tests, python-tests]
+    needs: [build, library-tests]
     steps:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -516,16 +513,16 @@ jobs:
           crate: cargo-nextest
           locked: true
       - name: KZG )tests
-        run: cargo nextest run --release --verbose tests_aggr::kzg_aggr_prove_and_verify_  --test-threads 8 -- --include-ignored
+        run: cargo nextest run --release --verbose tests_aggr::kzg_aggr_prove_and_verify_  --test-threads 4 -- --include-ignored
 
   prove-and-verify-aggr-evm-tests:
     runs-on: large-self-hosted
-    needs: [build, library-tests, python-tests]
+    needs: [build, library-tests]
     steps:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -535,7 +532,7 @@ jobs:
       - name: Install solc
         run: (hash svm 2>/dev/null || cargo install svm-rs) && svm install 0.8.20 && solc --version
       - name: Install Anvil
-        run: cargo install --git https://github.com/foundry-rs/foundry --rev 95a93cd397f25f3f8d49d2851eb52bc2d52dd983 --profile local --locked anvil --force
+        run: cargo install --git https://github.com/foundry-rs/foundry --rev b320f350156a0fb15c2eb13dc380deb2367c4474 --profile local --locked anvil --force
       - name: KZG prove and verify aggr tests
         run: cargo nextest run  --release --verbose tests_evm::kzg_evm_aggr_prove_and_verify_::t --test-threads 4 -- --include-ignored
 
@@ -546,7 +543,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -568,7 +565,7 @@ jobs:
           python-version: "3.7"
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - name: Install solc
@@ -576,7 +573,7 @@ jobs:
       - name: Setup Virtual Env and Install python dependencies
         run: python -m venv .env; source .env/bin/activate; pip install -r requirements.txt;
       - name: Install Anvil
-        run: cargo install --git https://github.com/foundry-rs/foundry --rev 95a93cd397f25f3f8d49d2851eb52bc2d52dd983 --profile local --locked anvil --force
+        run: cargo install --git https://github.com/foundry-rs/foundry --rev b320f350156a0fb15c2eb13dc380deb2367c4474 --profile local --locked anvil --force
       - name: Build python ezkl
         run: source .env/bin/activate; maturin develop --features python-bindings --release
       - name: Run pytest
@@ -592,7 +589,7 @@ jobs:
           python-version: "3.7"
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -623,7 +620,7 @@ jobs:
           python-version: "3.9"
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: baptiste0928/cargo-install@v1
@@ -633,7 +630,7 @@ jobs:
       - name: Install solc
         run: (hash svm 2>/dev/null || cargo install svm-rs) && svm install 0.8.20 && solc --version
       - name: Install Anvil
-        run: cargo install --git https://github.com/foundry-rs/foundry --rev 95a93cd397f25f3f8d49d2851eb52bc2d52dd983 --profile local --locked anvil --force
+        run: cargo install --git https://github.com/foundry-rs/foundry --rev b320f350156a0fb15c2eb13dc380deb2367c4474 --profile local --locked anvil --force
       - name: Setup Virtual Env and Install python dependencies
         run: python -m venv .env; source .env/bin/activate; pip install -r requirements.txt;
       - name: Build python ezkl

.github/workflows/wasm.yml

@@ -22,18 +22,15 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions-rs/toolchain@v1
         with:
-          toolchain: nightly-2023-08-24
+          toolchain: nightly-2024-01-04
           override: true
           components: rustfmt, clippy
       - uses: jetli/wasm-pack-action@v0.4.0
       - 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-2023-08-24-x86_64-unknown-linux-gnu
+        run: rustup component add rust-src --toolchain nightly-2024-01-04-x86_64-unknown-linux-gnu
       - name: Install binaryen
         run: |
             set -e

Cargo.toml

@@ -17,7 +17,7 @@ crate-type = ["cdylib", "rlib"]
 [dependencies]
 halo2_gadgets = { git = "https://github.com/zkonduit/halo2", branch= "main" }
 halo2_proofs = { git = "https://github.com/zkonduit/halo2", branch= "main"  }
-halo2curves = { version = "0.6.0", features = ["derive_serde"] }
+halo2curves = { git = "https://github.com/privacy-scaling-explorations/halo2curves", rev="9fff22c", features=["derive_serde"] }
 rand = { version = "0.8", default_features = false }
 itertools = { version = "0.10.3", default_features = false }
 clap = { version = "4.3.3", features = ["derive"]}
@@ -34,10 +34,13 @@ 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"
+tosubcommand = { git =  "https://github.com/zkonduit/enum_to_subcommand", package = "tosubcommand" }
+
 
 # evm related deps
 [target.'cfg(not(target_arch = "wasm32"))'.dependencies]
-ethers = { version = "2.0.7", default_features = false, features = ["ethers-solc"] }
+ethers = { version = "2.0.11", default_features = false, features = ["ethers-solc"] }
 indicatif = {version = "0.17.5", features = ["rayon"]}
 gag =  { version = "1.0.0", default_features = false}
 instant = { version = "0.1" }
@@ -158,6 +161,7 @@ mv-lookup = ["halo2_proofs/mv-lookup", "snark-verifier/mv-lookup", "halo2_solidi
 det-prove = []
 icicle = ["halo2_proofs/icicle_gpu"]
 empty-cmd = []
+no-banner = []
 
 # icicle patch to 0.1.0 if feature icicle is enabled
 [patch.'https://github.com/ingonyama-zk/icicle']

examples/conv2d_mnist/main.rs

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

examples/notebooks/data_attest_hashed.ipynb

@@ -309,7 +309,7 @@
             "metadata": {},
             "outputs": [],
             "source": [
-                "print(ezkl.string_to_felt(res['processed_outputs']['poseidon_hash'][0]))"
+                "print(ezkl.felt_to_big_endian(res['processed_outputs']['poseidon_hash'][0]))"
             ]
         },
         {
@@ -325,7 +325,7 @@
             "metadata": {},
             "outputs": [],
             "source": [
-                "from web3 import Web3, HTTPProvider, utils\n",
+                "from web3 import Web3, HTTPProvider\n",
                 "from solcx import compile_standard\n",
                 "from decimal import Decimal\n",
                 "import json\n",
@@ -338,7 +338,7 @@
                 "\n",
                 "def test_on_chain_data(res):\n",
                 "    # Step 0: Convert the tensor to a flat list\n",
-                "    data = [int(ezkl.string_to_felt(res['processed_outputs']['poseidon_hash'][0]), 0)]\n",
+                "    data = [int(ezkl.felt_to_big_endian(res['processed_outputs']['poseidon_hash'][0]), 0)]\n",
                 "\n",
                 "    # Step 1: Prepare the data\n",
                 "    # Step 2: Prepare and compile the contract.\n",
@@ -648,7 +648,7 @@
             "name": "python",
             "nbconvert_exporter": "python",
             "pygments_lexer": "ipython3",
-            "version": "3.9.13"
+            "version": "3.9.15"
         },
         "orig_nbformat": 4
     },

examples/notebooks/ezkl_demo.ipynb

@@ -695,7 +695,7 @@
         "formatted_output = \"[\"\n",
         "for i, value in enumerate(proof[\"instances\"]):\n",
         "    for j, field_element in enumerate(value):\n",
-        "        onchain_input_array.append(ezkl.string_to_felt(field_element))\n",
+        "        onchain_input_array.append(ezkl.felt_to_big_endian(field_element))\n",
         "        formatted_output += str(onchain_input_array[-1])\n",
         "        if j != len(value) - 1:\n",
         "            formatted_output += \", \"\n",
@@ -705,7 +705,7 @@
         "# copy them over to remix and see if they verify\n",
         "# What happens when you change a value?\n",
         "print(\"pubInputs: \", formatted_output)\n",
-        "print(\"proof: \", \"0x\" + proof[\"proof\"])"
+        "print(\"proof: \", proof[\"proof\"])"
       ]
     },
     {

examples/notebooks/set_membership.ipynb

@@ -122,8 +122,8 @@
         "# Loop through each element in the y tensor\n",
         "for e in y_input:\n",
         "    # Apply the custom function and append the result to the list\n",
-        "    print(ezkl.float_to_string(e,7))\n",
-        "    result.append(ezkl.poseidon_hash([ezkl.float_to_string(e, 7)])[0])\n",
+        "    print(ezkl.float_to_felt(e,7))\n",
+        "    result.append(ezkl.poseidon_hash([ezkl.float_to_felt(e, 7)])[0])\n",
         "\n",
         "y = y.unsqueeze(0)\n",
         "y = y.reshape(1, 9)\n",

examples/notebooks/solvency.ipynb

@@ -126,7 +126,7 @@
         "# Loop through each element in the y tensor\n",
         "for e in user_preimages:\n",
         "    # Apply the custom function and append the result to the list\n",
-        "    users.append(ezkl.poseidon_hash([ezkl.float_to_string(e, 0)])[0])\n",
+        "    users.append(ezkl.poseidon_hash([ezkl.float_to_felt(e, 0)])[0])\n",
         "\n",
         "users_t = torch.tensor(user_preimages)\n",
         "users_t = users_t.reshape(1, 6)\n",
@@ -303,7 +303,7 @@
         "# we force the output to be 1 this corresponds to the solvency test being true -- and we set this to a fixed vis output\n",
         "# this means that the output is fixed and the verifier can see it but that if the input is not in the set the output will not be 0 and the verifier will reject\n",
         "witness = json.load(open(witness_path, \"r\"))\n",
-        "witness[\"outputs\"][0] = [ezkl.float_to_string(1.0, 0)]\n",
+        "witness[\"outputs\"][0] = [ezkl.float_to_felt(1.0, 0)]\n",
         "json.dump(witness, open(witness_path, \"w\"))"
       ]
     },
@@ -417,7 +417,7 @@
         "# we force the output to be 1 this corresponds to the solvency test being true -- and we set this to a fixed vis output\n",
         "# this means that the output is fixed and the verifier can see it but that if the input is not in the set the output will not be 0 and the verifier will reject\n",
         "witness = json.load(open(witness_path, \"r\"))\n",
-        "witness[\"outputs\"][0] = [ezkl.float_to_string(1.0, 0)]\n",
+        "witness[\"outputs\"][0] = [ezkl.float_to_felt(1.0, 0)]\n",
         "json.dump(witness, open(witness_path, \"w\"))\n"
       ]
     },
@@ -510,7 +510,7 @@
       "name": "python",
       "nbconvert_exporter": "python",
       "pygments_lexer": "ipython3",
-      "version": "3.9.13"
+      "version": "3.9.15"
     }
   },
   "nbformat": 4,

examples/notebooks/tictactoe_autoencoder.ipynb

@@ -633,7 +633,7 @@
                 "json.dump(data, open(cal_path, 'w'))\n",
                 "\n",
                 "\n",
-                "ezkl.calibrate_settings(cal_path, model_path, settings_path, \"resources\", scales = [4])"
+                "ezkl.calibrate_settings(cal_path, model_path, settings_path, \"resources\", scales = [11])"
             ]
         },
         {
@@ -664,7 +664,6 @@
                 "    compiled_model_path,\n",
                 "    vk_path,\n",
                 "    pk_path,\n",
-                "    \n",
                 ")"
             ]
         },

examples/notebooks/world_rotation.ipynb

@@ -503,11 +503,11 @@
     "pyplot.arrow(0, 0, 1, 0, width=0.02, alpha=0.5)\n",
     "pyplot.arrow(0, 0, 0, 1, width=0.02, alpha=0.5)\n",
     "\n",
-    "arrow_x = ezkl.string_to_float(witness['outputs'][0][0], out_scale)\n",
-    "arrow_y = ezkl.string_to_float(witness['outputs'][0][1], out_scale)\n",
+    "arrow_x = ezkl.felt_to_float(witness['outputs'][0][0], out_scale)\n",
+    "arrow_y = ezkl.felt_to_float(witness['outputs'][0][1], out_scale)\n",
     "pyplot.arrow(0, 0, arrow_x, arrow_y, width=0.02)\n",
-    "arrow_x = ezkl.string_to_float(witness['outputs'][0][2], out_scale)\n",
-    "arrow_y = ezkl.string_to_float(witness['outputs'][0][3], out_scale)\n",
+    "arrow_x = ezkl.felt_to_float(witness['outputs'][0][2], out_scale)\n",
+    "arrow_y = ezkl.felt_to_float(witness['outputs'][0][3], out_scale)\n",
     "pyplot.arrow(0, 0, arrow_x, arrow_y, width=0.02)"
    ]
   }

src/bin/ezkl.rs

@@ -11,8 +11,8 @@ use ezkl::execute::run;
 #[cfg(not(target_arch = "wasm32"))]
 use ezkl::logger::init_logger;
 #[cfg(not(target_arch = "wasm32"))]
-use log::{error, info};
-#[cfg(not(target_arch = "wasm32"))]
+use log::{debug, error, info};
+#[cfg(not(any(target_arch = "wasm32", feature = "no-banner")))]
 use rand::prelude::SliceRandom;
 #[cfg(not(target_arch = "wasm32"))]
 #[cfg(feature = "icicle")]
@@ -25,6 +25,7 @@ use std::error::Error;
 pub async fn main() -> Result<(), Box<dyn Error>> {
     let args = Cli::parse();
     init_logger();
+    #[cfg(not(any(target_arch = "wasm32", feature = "no-banner")))]
     banner();
     #[cfg(feature = "icicle")]
     if env::var("ENABLE_ICICLE_GPU").is_ok() {
@@ -32,7 +33,7 @@ pub async fn main() -> Result<(), Box<dyn Error>> {
     } else {
         info!("Running with CPU");
     }
-    info!("command: \n {}", &args.as_json()?.to_colored_json_auto()?);
+    debug!("command: \n {}", &args.as_json()?.to_colored_json_auto()?);
     let res = run(args.command).await;
     match &res {
         Ok(_) => info!("succeeded"),
@@ -44,7 +45,7 @@ pub async fn main() -> Result<(), Box<dyn Error>> {
 #[cfg(target_arch = "wasm32")]
 pub fn main() {}
 
-#[cfg(not(target_arch = "wasm32"))]
+#[cfg(not(any(target_arch = "wasm32", feature = "no-banner")))]
 fn banner() {
     let ell: Vec<&str> = vec![
         "for Neural Networks",

src/circuit/modules/kzg.rs

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

src/circuit/modules/planner.rs

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

src/circuit/ops/base.rs

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

src/circuit/ops/chip.rs

@@ -16,10 +16,11 @@ use pyo3::{
     types::PyString,
 };
 use serde::{Deserialize, Serialize};
+use tosubcommand::ToFlags;
 
 use crate::{
-    circuit::ops::base::BaseOp,
     circuit::{
+        ops::base::BaseOp,
         table::{Range, RangeCheck, Table},
         utils,
     },
@@ -61,6 +62,22 @@ pub enum CheckMode {
     UNSAFE,
 }
 
+impl std::fmt::Display for CheckMode {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        match self {
+            CheckMode::SAFE => write!(f, "safe"),
+            CheckMode::UNSAFE => write!(f, "unsafe"),
+        }
+    }
+}
+
+impl ToFlags for CheckMode {
+    /// Convert the struct to a subcommand string
+    fn to_flags(&self) -> Vec<String> {
+        vec![format!("{}", self)]
+    }
+}
+
 impl From<String> for CheckMode {
     fn from(value: String) -> Self {
         match value.to_lowercase().as_str() {
@@ -83,6 +100,19 @@ pub struct Tolerance {
     pub scale: utils::F32,
 }
 
+impl std::fmt::Display for Tolerance {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(f, "{:.2}", self.val)
+    }
+}
+
+impl ToFlags for Tolerance {
+    /// Convert the struct to a subcommand string
+    fn to_flags(&self) -> Vec<String> {
+        vec![format!("{}", self)]
+    }
+}
+
 impl FromStr for Tolerance {
     type Err = String;
 
@@ -276,9 +306,20 @@ impl<F: PrimeField + TensorType + PartialOrd> BaseConfig<F> {
 
                 let constraints = match base_op {
                     BaseOp::IsBoolean => {
-                        vec![(qis[1].clone()) * (qis[1].clone() - Expression::Constant(F::from(1)))]
+                        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()]
                     }
-                    BaseOp::IsZero => vec![qis[1].clone()],
                     _ => {
                         let expected_output: Tensor<Expression<F>> = output
                             .query_rng(meta, *block_idx, *inner_col_idx, rotation_offset, rng)
@@ -499,7 +540,9 @@ impl<F: PrimeField + TensorType + PartialOrd> BaseConfig<F> {
         &mut self,
         cs: &mut ConstraintSystem<F>,
         input: &VarTensor,
+        index: &VarTensor,
         range: Range,
+        logrows: usize,
     ) -> Result<(), Box<dyn Error>>
     where
         F: Field,
@@ -512,43 +555,72 @@ impl<F: PrimeField + TensorType + PartialOrd> BaseConfig<F> {
 
         // we borrow mutably twice so we need to do this dance
 
-        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);
-            self.range_checks.insert(range, range_check.clone());
-            range_check
-        } else {
-            return Ok(());
-        };
+        let range_check =
+            if let std::collections::btree_map::Entry::Vacant(e) = self.range_checks.entry(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, logrows);
+                e.insert(range_check.clone());
+                range_check
+            } else {
+                return Ok(());
+            };
 
         for x in 0..input.num_blocks() {
             for y in 0..input.num_inner_cols() {
-                let single_col_sel = cs.complex_selector();
+                let len = range_check.selector_constructor.degree;
+                let multi_col_selector = cs.complex_selector();
 
-                cs.lookup("", |cs| {
-                    let mut res = vec![];
-                    let sel = cs.query_selector(single_col_sel);
+                for (col_idx, input_col) in range_check.inputs.iter().enumerate() {
+                    cs.lookup("", |cs| {
+                        let mut res = vec![];
+                        let sel = cs.query_selector(multi_col_selector);
 
-                    let input_query = match &input {
-                        VarTensor::Advice { inner: advices, .. } => {
-                            cs.query_advice(advices[x][y], Rotation(0))
-                        }
-                        _ => unreachable!(),
-                    };
+                        let synthetic_sel = match len {
+                            1 => Expression::Constant(F::from(1)),
+                            _ => match index {
+                                VarTensor::Advice { inner: advices, .. } => {
+                                    cs.query_advice(advices[x][y], Rotation(0))
+                                }
+                                _ => unreachable!(),
+                            },
+                        };
 
-                    let default_x = range_check.get_first_element();
+                        let input_query = match &input {
+                            VarTensor::Advice { inner: advices, .. } => {
+                                cs.query_advice(advices[x][y], Rotation(0))
+                            }
+                            _ => unreachable!(),
+                        };
 
-                    let not_sel = Expression::Constant(F::ONE) - sel.clone();
+                        let default_x = range_check.get_first_element(col_idx);
 
-                    res.extend([(
-                        sel.clone() * input_query.clone()
-                            + not_sel.clone() * Expression::Constant(default_x),
-                        range_check.input,
-                    )]);
+                        let col_expr = sel.clone()
+                            * range_check
+                                .selector_constructor
+                                .get_expr_at_idx(col_idx, synthetic_sel);
 
-                    res
-                });
-                selectors.insert((range, x, y), single_col_sel);
+                        let multiplier = range_check
+                            .selector_constructor
+                            .get_selector_val_at_idx(col_idx);
+
+                        let not_expr = Expression::Constant(multiplier) - col_expr.clone();
+
+                        res.extend([(
+                            col_expr.clone() * input_query.clone()
+                                + not_expr.clone() * Expression::Constant(default_x),
+                            *input_col,
+                        )]);
+
+                        log::trace!("---------------- col {:?} ------------------", col_idx,);
+                        log::trace!("expr: {:?}", col_expr,);
+                        log::trace!("multiplier: {:?}", multiplier);
+                        log::trace!("not_expr: {:?}", not_expr);
+                        log::trace!("default x: {:?}", default_x);
+
+                        res
+                    });
+                }
+                selectors.insert((range, x, y), multi_col_selector);
             }
         }
         self.range_check_selectors.extend(selectors);
@@ -558,6 +630,11 @@ impl<F: PrimeField + TensorType + PartialOrd> BaseConfig<F> {
             self.lookup_input = input.clone();
         }
 
+        if let VarTensor::Empty = self.lookup_index {
+            debug!("assigning lookup index");
+            self.lookup_index = index.clone();
+        }
+
         Ok(())
     }
 

src/circuit/ops/hybrid.rs

@@ -6,7 +6,6 @@ use crate::{
     tensor::{self, Tensor, TensorError, TensorType, ValTensor},
 };
 use halo2curves::ff::PrimeField;
-use itertools::Itertools;
 use serde::{Deserialize, Serialize};
 // import run args from model
 
@@ -69,14 +68,6 @@ pub enum HybridOp {
         dim: usize,
         num_classes: usize,
     },
-    GatherElements {
-        dim: usize,
-        constant_idx: Option<Tensor<usize>>,
-    },
-    ScatterElements {
-        dim: usize,
-        constant_idx: Option<Tensor<usize>>,
-    },
 }
 
 impl<F: PrimeField + TensorType + PartialOrd> Op<F> for HybridOp {
@@ -84,7 +75,6 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for HybridOp {
     fn requires_homogenous_input_scales(&self) -> Vec<usize> {
         match self {
             HybridOp::Greater | HybridOp::Less | HybridOp::Equals => vec![0, 1],
-            HybridOp::ScatterElements { .. } => vec![0, 2],
             HybridOp::GreaterEqual | HybridOp::LessEqual => vec![0, 1],
             _ => vec![],
         }
@@ -98,176 +88,42 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for HybridOp {
     fn f(&self, inputs: &[Tensor<F>]) -> Result<ForwardResult<F>, TensorError> {
         let x = inputs[0].clone().map(|x| felt_to_i128(x));
 
-        let (res, intermediate_lookups) = match &self {
-            HybridOp::ReduceMax { axes, .. } => {
-                let res = tensor::ops::max_axes(&x, axes)?;
-                let max_minus_one =
-                    Tensor::from(vec![x.clone().into_iter().max().unwrap() - 1].into_iter());
-                let unit = Tensor::from(vec![1].into_iter());
-                // relu(x - max(x - 1)
-                let inter_1 = (x.clone() - max_minus_one)?;
-                // relu(1 - sum(relu(inter_1)))
-                let inter_2 = (unit
-                    - tensor::ops::sum(&tensor::ops::nonlinearities::leakyrelu(&inter_1, 0.0))?)?;
-
-                (res.clone(), vec![inter_1, inter_2])
-            }
-            HybridOp::ReduceMin { axes, .. } => {
-                let res = tensor::ops::min_axes(&x, axes)?;
-                let min_plus_one =
-                    Tensor::from(vec![x.clone().into_iter().min().unwrap() + 1].into_iter());
-                let unit = Tensor::from(vec![1].into_iter());
-                // relu(min(x + 1) - x)
-                let inter_1 = (min_plus_one - x.clone())?;
-                // relu(1 - sum(relu(inter_1)))
-                let inter_2 = (unit
-                    - tensor::ops::sum(&tensor::ops::nonlinearities::leakyrelu(&inter_1, 0.0))?)?;
-                (res.clone(), vec![inter_1, inter_2])
-            }
-            HybridOp::Div {
-                denom,
-                use_range_check_for_int,
-                ..
-            } => {
-                let res = crate::tensor::ops::nonlinearities::const_div(&x, denom.0 as f64);
-                // if denom is a round number and use_range_check_for_int is true, use range check check
-                if denom.0.fract() == 0.0 && *use_range_check_for_int {
-                    let divisor = Tensor::from(vec![denom.0 as i128 / 2].into_iter());
-                    (res, vec![-divisor.clone(), divisor])
-                } else {
-                    (res, vec![x])
-                }
+        let res = match &self {
+            HybridOp::ReduceMax { axes, .. } => tensor::ops::max_axes(&x, axes)?,
+            HybridOp::ReduceMin { axes, .. } => tensor::ops::min_axes(&x, axes)?,
+            HybridOp::Div { denom, .. } => {
+                crate::tensor::ops::nonlinearities::const_div(&x, denom.0 as f64)
             }
             HybridOp::Recip {
                 input_scale,
                 output_scale,
-                use_range_check_for_int,
-            } => {
-                let res = crate::tensor::ops::nonlinearities::recip(
-                    &x,
-                    input_scale.0 as f64,
-                    output_scale.0 as f64,
-                );
-                // if scale is a round number and use_range_check_for_int is true, use range check check
-                if input_scale.0.fract() == 0.0 && *use_range_check_for_int {
-                    let err_tol = Tensor::from(
-                        vec![(output_scale.0 * input_scale.0) as i128 / 2].into_iter(),
-                    );
-                    (res, vec![-err_tol.clone(), err_tol])
-                } else {
-                    (res, vec![x])
-                }
-            }
-            HybridOp::ReduceArgMax { dim } => {
-                let res = tensor::ops::argmax_axes(&x, *dim)?;
-                let 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)
-            }
+                ..
+            } => crate::tensor::ops::nonlinearities::recip(
+                &x,
+                input_scale.0 as f64,
+                output_scale.0 as f64,
+            ),
+            HybridOp::ReduceArgMax { dim } => tensor::ops::argmax_axes(&x, *dim)?,
+            HybridOp::ReduceArgMin { dim } => tensor::ops::argmin_axes(&x, *dim)?,
             HybridOp::Gather { dim, constant_idx } => {
                 if let Some(idx) = constant_idx {
-                    log::debug!("idx: {}", idx.show());
-                    let res = tensor::ops::gather(&x, idx, *dim)?;
-                    (res.clone(), vec![])
+                    tensor::ops::gather(&x, idx, *dim)?
                 } else {
                     let y = inputs[1].clone().map(|x| felt_to_i128(x));
-                    let 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)
+                    tensor::ops::gather(&x, &y.map(|x| x as usize), *dim)?
                 }
             }
             HybridOp::OneHot { dim, num_classes } => {
-                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)
+                tensor::ops::one_hot(&x, *num_classes, *dim)?.clone()
             }
-            HybridOp::TopK { dim, k, largest } => {
-                let res = tensor::ops::topk_axes(&x, *k, *dim, *largest)?;
 
-                let mut inter_equals = x
-                    .clone()
-                    .into_iter()
-                    .flat_map(|elem| {
-                        tensor::ops::equals(&res, &vec![elem].into_iter().into())
-                            .unwrap()
-                            .1
-                    })
-                    .collect::<Vec<_>>();
-
-                // sort in descending order and take pairwise differences
-                inter_equals.push(
-                    x.into_iter()
-                        .sorted()
-                        .tuple_windows()
-                        .map(|(a, b)| b - a)
-                        .into(),
-                );
-
-                (res.clone(), inter_equals)
-            }
-            HybridOp::GatherElements { dim, constant_idx } => {
-                if let Some(idx) = constant_idx {
-                    log::debug!("idx: {}", idx.show());
-                    let res = tensor::ops::gather_elements(&x, idx, *dim)?;
-                    (res.clone(), vec![])
-                } else {
-                    let y = inputs[1].clone().map(|x| felt_to_i128(x));
-                    let 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::TopK { dim, k, largest } => tensor::ops::topk_axes(&x, *k, *dim, *largest)?,
             HybridOp::MaxPool2d {
                 padding,
                 stride,
                 pool_dims,
                 ..
-            } => {
-                let max_minus_one =
-                    Tensor::from(vec![x.clone().into_iter().max().unwrap() - 1].into_iter());
-                let unit = Tensor::from(vec![1].into_iter());
-                // relu(x - max(x - 1)
-                let inter_1 = (x.clone() - max_minus_one)?;
-                // relu(1 - sum(relu(inter_1)))
-                let inter_2 = (unit
-                    - tensor::ops::sum(&tensor::ops::nonlinearities::leakyrelu(&inter_1, 0.0))?)?;
-                (
-                    tensor::ops::max_pool2d(&x, padding, stride, pool_dims)?,
-                    vec![inter_1, inter_2],
-                )
-            }
+            } => tensor::ops::max_pool2d(&x, padding, stride, pool_dims)?,
             HybridOp::SumPool {
                 padding,
                 stride,
@@ -279,10 +135,7 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for HybridOp {
             }
             HybridOp::RangeCheck(tol) => {
                 let y = inputs[1].clone().map(|x| felt_to_i128(x));
-                (
-                    tensor::ops::nonlinearities::range_check_percent(&[x, y], 128, 128, tol.val),
-                    vec![],
-                )
+                tensor::ops::nonlinearities::range_check_percent(&[x, y], 128, 128, tol.val)
             }
             HybridOp::Greater => {
                 let y = inputs[1].clone().map(|x| felt_to_i128(x));
@@ -309,10 +162,7 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for HybridOp {
         // convert back to felt
         let output = res.map(|x| i128_to_felt(x));
 
-        Ok(ForwardResult {
-            output,
-            intermediate_lookups,
-        })
+        Ok(ForwardResult { output })
     }
 
     fn as_string(&self) -> String {
@@ -366,8 +216,6 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for HybridOp {
             HybridOp::TopK { k, dim, largest } => {
                 format!("TOPK (k={}, dim={}, largest={})", k, dim, largest)
             }
-            HybridOp::GatherElements { dim, .. } => format!("GATHERELEMENTS (dim={})", dim),
-            HybridOp::ScatterElements { dim, .. } => format!("SCATTERELEMENTS (dim={})", dim),
             HybridOp::OneHot { dim, num_classes } => {
                 format!("ONEHOT (dim={}, num_classes={})", dim, num_classes)
             }
@@ -429,7 +277,7 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for HybridOp {
                 ..
             } => {
                 if denom.0.fract() == 0.0 && *use_range_check_for_int {
-                    layouts::div(
+                    layouts::loop_div(
                         config,
                         region,
                         values[..].try_into()?,
@@ -440,9 +288,7 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for HybridOp {
                         config,
                         region,
                         values.try_into()?,
-                        &LookupOp::Div {
-                            denom: denom.clone(),
-                        },
+                        &LookupOp::Div { denom: *denom },
                     )?
                 }
             }
@@ -453,26 +299,7 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for HybridOp {
                     layouts::gather(config, region, values[..].try_into()?, *dim)?
                 }
             }
-            HybridOp::GatherElements { dim, constant_idx } => {
-                if let Some(idx) = constant_idx {
-                    tensor::ops::gather_elements(values[0].get_inner_tensor()?, idx, *dim)?.into()
-                } else {
-                    layouts::gather_elements(config, region, values[..].try_into()?, *dim)?
-                }
-            }
-            HybridOp::ScatterElements { dim, constant_idx } => {
-                if let Some(idx) = constant_idx {
-                    tensor::ops::scatter(
-                        values[0].get_inner_tensor()?,
-                        idx,
-                        values[1].get_inner_tensor()?,
-                        *dim,
-                    )?
-                    .into()
-                } else {
-                    layouts::scatter_elements(config, region, values[..].try_into()?, *dim)?
-                }
-            }
+
             HybridOp::MaxPool2d {
                 padding,
                 stride,

src/circuit/ops/layouts.rs

@@ -18,10 +18,7 @@ use super::{
     region::RegionCtx,
 };
 use crate::{
-    circuit::{
-        ops::base::BaseOp,
-        utils::{self},
-    },
+    circuit::{ops::base::BaseOp, utils},
     fieldutils::{felt_to_i128, i128_to_felt},
     tensor::{
         get_broadcasted_shape,
@@ -54,6 +51,41 @@ pub fn overflowed_len(starting_idx: usize, mut total_len: usize, column_len: usi
     total_len
 }
 
+/// Same as div but splits the division into N parts
+pub fn loop_div<F: PrimeField + TensorType + PartialOrd>(
+    config: &BaseConfig<F>,
+    region: &mut RegionCtx<F>,
+    value: &[ValTensor<F>; 1],
+    divisor: F,
+) -> Result<ValTensor<F>, Box<dyn Error>> {
+    if divisor == F::ONE {
+        return Ok(value[0].clone());
+    }
+
+    // if integer val is divisible by 2, we can use a faster method and div > F::S
+    let mut divisor = divisor;
+    let mut num_parts = 1;
+
+    while felt_to_i128(divisor) % 2 == 0 && felt_to_i128(divisor) > (2_i128.pow(F::S - 4)) {
+        divisor = i128_to_felt(felt_to_i128(divisor) / 2);
+        num_parts += 1;
+    }
+
+    let output = div(config, region, value, divisor)?;
+    if num_parts == 1 {
+        return Ok(output);
+    }
+
+    let divisor_int = 2_i128.pow(num_parts - 1);
+    let divisor_felt = i128_to_felt(divisor_int);
+    if divisor_int <= 2_i128.pow(F::S - 3) {
+        div(config, region, &[output], divisor_felt)
+    } else {
+        // keep splitting the divisor until it satisfies the condition
+        loop_div(config, region, &[output], divisor_felt)
+    }
+}
+
 /// Div accumulated layout
 pub fn div<F: PrimeField + TensorType + PartialOrd>(
     config: &BaseConfig<F>,
@@ -61,6 +93,10 @@ pub fn div<F: PrimeField + TensorType + PartialOrd>(
     value: &[ValTensor<F>; 1],
     div: F,
 ) -> Result<ValTensor<F>, Box<dyn Error>> {
+    if div == F::ONE {
+        return Ok(value[0].clone());
+    }
+
     let input = value[0].clone();
     let input_dims = input.dims();
 
@@ -88,6 +124,8 @@ pub fn div<F: PrimeField + TensorType + PartialOrd>(
         .into()
     };
     claimed_output.reshape(input_dims)?;
+    region.assign(&config.output, &claimed_output)?;
+    region.increment(claimed_output.len());
 
     let product = pairwise(
         config,
@@ -96,8 +134,6 @@ pub fn div<F: PrimeField + TensorType + PartialOrd>(
         BaseOp::Mult,
     )?;
 
-    log::debug!("product: {:?}", product.get_int_evals()?);
-
     let diff_with_input = pairwise(
         config,
         region,
@@ -115,6 +151,46 @@ pub fn div<F: PrimeField + TensorType + PartialOrd>(
     Ok(claimed_output)
 }
 
+fn recip_int<F: PrimeField + TensorType + PartialOrd>(
+    config: &BaseConfig<F>,
+    region: &mut RegionCtx<F>,
+    input: &[ValTensor<F>; 1],
+) -> Result<ValTensor<F>, Box<dyn Error>> {
+    // assert is boolean
+    let zero_inverse_val = tensor::ops::nonlinearities::zero_recip(1.0)[0];
+    // get values where input is 0
+    let zero_mask = equals_zero(config, region, input)?;
+
+    let one_minus_zero_mask = pairwise(
+        config,
+        region,
+        &[
+            zero_mask.clone(),
+            ValTensor::from(Tensor::from([ValType::Constant(F::ONE)].into_iter())),
+        ],
+        BaseOp::Sub,
+    )?;
+
+    let zero_inverse_val = pairwise(
+        config,
+        region,
+        &[
+            zero_mask,
+            ValTensor::from(Tensor::from(
+                [ValType::Constant(i128_to_felt(zero_inverse_val))].into_iter(),
+            )),
+        ],
+        BaseOp::Mult,
+    )?;
+
+    pairwise(
+        config,
+        region,
+        &[one_minus_zero_mask, zero_inverse_val],
+        BaseOp::Add,
+    )
+}
+
 /// recip accumulated layout
 pub fn recip<F: PrimeField + TensorType + PartialOrd>(
     config: &BaseConfig<F>,
@@ -123,18 +199,25 @@ pub fn recip<F: PrimeField + TensorType + PartialOrd>(
     input_scale: F,
     output_scale: F,
 ) -> Result<ValTensor<F>, Box<dyn Error>> {
+    if output_scale == F::ONE || output_scale == F::ZERO {
+        return recip_int(config, region, value);
+    }
+
     let input = value[0].clone();
     let input_dims = input.dims();
 
-    let range_check_bracket = felt_to_i128(output_scale * input_scale) / 2;
+    let integer_input_scale = felt_to_i128(input_scale);
+    let integer_output_scale = felt_to_i128(output_scale);
 
-    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());
+    // range_check_bracket is min of input_scale * output_scale and 2^F::S - 3
+    let range_check_len = std::cmp::min(integer_output_scale, 2_i128.pow(F::S - 4));
 
-    let is_assigned = !input.any_unknowns()? && !scaled_unit.any_unknowns()?;
+    let input_scale_ratio =
+        i128_to_felt(integer_input_scale * integer_output_scale / range_check_len);
+
+    let range_check_bracket = range_check_len / 2;
+
+    let is_assigned = !input.any_unknowns()?;
 
     let mut claimed_output: ValTensor<F> = if is_assigned {
         let input_evals = input.get_int_evals()?;
@@ -155,6 +238,8 @@ pub fn recip<F: PrimeField + TensorType + PartialOrd>(
         .into()
     };
     claimed_output.reshape(input_dims)?;
+    let claimed_output = region.assign(&config.output, &claimed_output)?;
+    region.increment(claimed_output.len());
 
     // this is now of scale 2 * scale
     let product = pairwise(
@@ -164,29 +249,47 @@ pub fn recip<F: PrimeField + TensorType + PartialOrd>(
         BaseOp::Mult,
     )?;
 
-    log::debug!("product: {:?}", product.get_int_evals()?);
+    // divide by input_scale
+    let rebased_div = loop_div(config, region, &[product], input_scale_ratio)?;
 
-    // this is now of scale 2 * scale hence why we rescaled the unit scale
-    let diff_with_input = pairwise(
+    let zero_inverse_val =
+        tensor::ops::nonlinearities::zero_recip(felt_to_i128(output_scale) as f64)[0];
+    let zero_inverse =
+        Tensor::from([ValType::Constant(i128_to_felt::<F>(zero_inverse_val))].into_iter());
+
+    let equal_zero_mask = equals_zero(config, region, &[input.clone()])?;
+
+    let equal_inverse_mask = equals(
         config,
         region,
-        &[product.clone(), scaled_unit.clone()],
-        BaseOp::Sub,
+        &[claimed_output.clone(), zero_inverse.into()],
     )?;
 
-    log::debug!("scaled_unit: {:?}", scaled_unit.get_int_evals()?);
+    // assert the two masks are equal
+    enforce_equality(
+        config,
+        region,
+        &[equal_zero_mask.clone(), equal_inverse_mask],
+    )?;
 
-    // debug print the diff
-    log::debug!("diff_with_input: {:?}", diff_with_input.get_int_evals()?);
+    let unit_scale = Tensor::from([ValType::Constant(i128_to_felt(range_check_len))].into_iter());
 
-    log::debug!("range_check_bracket: {:?}", range_check_bracket);
+    let unit_mask = pairwise(
+        config,
+        region,
+        &[equal_zero_mask, unit_scale.into()],
+        BaseOp::Mult,
+    )?;
+
+    // now add the unit mask to the rebased_div
+    let rebased_offset_div = pairwise(config, region, &[rebased_div, unit_mask], BaseOp::Add)?;
 
     // at most the error should be in the original unit scale's range
     range_check(
         config,
         region,
-        &[diff_with_input],
-        &(-range_check_bracket, range_check_bracket),
+        &[rebased_offset_div],
+        &(range_check_bracket, 3 * range_check_bracket),
     )?;
 
     Ok(claimed_output)
@@ -233,7 +336,7 @@ pub fn dot<F: PrimeField + TensorType + PartialOrd>(
 
     let mut assigned_len = 0;
     for (i, input) in values.iter_mut().enumerate() {
-        input.pad_to_zero_rem(block_width)?;
+        input.pad_to_zero_rem(block_width, ValType::Constant(F::ZERO))?;
         let inp = {
             let (res, len) = region.assign_with_duplication(
                 &config.inputs[i],
@@ -789,14 +892,7 @@ 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 = 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)?;
-        }
-    }
+    let assigned_output = boolean_identity(config, region, &[output.clone()], true)?;
     region.increment(std::cmp::max(assigned_output.len(), assigned_input.len()));
 
     let sum = sum(config, region, &[assigned_output.clone()])?;
@@ -1168,7 +1264,7 @@ pub fn sum<F: PrimeField + TensorType + PartialOrd>(
     let assigned_len: usize;
     let input = {
         let mut input = values[0].clone();
-        input.pad_to_zero_rem(block_width)?;
+        input.pad_to_zero_rem(block_width, ValType::Constant(F::ZERO))?;
         let (res, len) =
             region.assign_with_duplication(&config.inputs[1], &input, &config.check_mode, false)?;
         assigned_len = len;
@@ -1237,7 +1333,7 @@ pub fn prod<F: PrimeField + TensorType + PartialOrd>(
     let assigned_len: usize;
     let input = {
         let mut input = values[0].clone();
-        input.pad_to_zero_rem(block_width)?;
+        input.pad_to_zero_rem(block_width, ValType::Constant(F::ONE))?;
         let (res, len) =
             region.assign_with_duplication(&config.inputs[1], &input, &config.check_mode, false)?;
         assigned_len = len;
@@ -1701,10 +1797,42 @@ 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)?;
+    equals_zero(config, region, &[diff])
+}
 
-    let res = nonlinearity(config, region, &[diff], &LookupOp::KroneckerDelta)?;
+/// Equality boolean operation
+pub fn equals_zero<F: PrimeField + TensorType + PartialOrd>(
+    config: &BaseConfig<F>,
+    region: &mut RegionCtx<F>,
+    values: &[ValTensor<F>; 1],
+) -> Result<ValTensor<F>, Box<dyn Error>> {
+    let values = values[0].clone();
+    let values_inverse = values.inverse()?;
+    let product_values_and_invert = pairwise(
+        config,
+        region,
+        &[values.clone(), values_inverse],
+        BaseOp::Mult,
+    )?;
 
-    Ok(res)
+    // constant of 1
+    let mut ones = Tensor::from(vec![ValType::Constant(F::from(1))].into_iter());
+    ones.set_visibility(&crate::graph::Visibility::Fixed);
+
+    // subtract
+    let output = pairwise(
+        config,
+        region,
+        &[ones.into(), product_values_and_invert],
+        BaseOp::Sub,
+    )?;
+
+    // take the product of diff and output
+    let prod_check = pairwise(config, region, &[values, output.clone()], BaseOp::Mult)?;
+
+    is_zero_identity(config, region, &[prod_check], false)?;
+
+    Ok(output)
 }
 
 /// Xor boolean operation
@@ -1768,21 +1896,7 @@ pub fn iff<F: PrimeField + TensorType + PartialOrd>(
             .into();
 
     // make sure mask is boolean
-    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 assigned_mask = boolean_identity(config, region, &[mask.clone()], true)?;
 
     let one_minus_mask = pairwise(config, region, &[unit, assigned_mask.clone()], BaseOp::Sub)?;
 
@@ -1876,17 +1990,15 @@ pub fn sumpool<F: PrimeField + TensorType + PartialOrd>(
     let shape = &res[0].dims()[2..];
     let mut last_elem = res[1..]
         .iter()
-        .fold(Ok(res[0].clone()), |acc, elem| acc?.concat(elem.clone()))?;
+        .try_fold(res[0].clone(), |acc, elem| acc.concat(elem.clone()))?;
     last_elem.reshape(&[&[batch_size, image_channels], shape].concat())?;
 
     if normalized {
-        last_elem = nonlinearity(
+        last_elem = loop_div(
             config,
             region,
             &[last_elem],
-            &LookupOp::Div {
-                denom: utils::F32((kernel_shape.0 * kernel_shape.1) as f32),
-            },
+            F::from((kernel_shape.0 * kernel_shape.1) as u64),
         )?;
     }
     Ok(last_elem)
@@ -2383,18 +2495,60 @@ 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 = region.assign(&config.inputs[1], &values[0])?;
+    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()
+    };
     // Enable the selectors
     if !region.is_dummy() {
         (0..output.len())
             .map(|j| {
-                let (x, y, z) = config.inputs[1].cartesian_coord(region.linear_coord() + j);
+                let index = region.linear_coord() - j - 1;
+
+                let (x, y, z) = config.output.cartesian_coord(index);
                 let selector = config.selectors.get(&(BaseOp::IsBoolean, x, y));
 
                 region.enable(selector, z)?;
@@ -2402,7 +2556,6 @@ pub fn boolean_identity<F: PrimeField + TensorType + PartialOrd>(
             })
             .collect::<Result<Vec<_>, Box<dyn Error>>>()?;
     }
-    region.increment(output.len());
 
     Ok(output)
 }
@@ -2452,7 +2605,7 @@ 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();
@@ -2465,19 +2618,52 @@ pub fn range_check<F: PrimeField + TensorType + PartialOrd>(
 
     let is_dummy = region.is_dummy();
 
+    let table_index: ValTensor<F> = w
+        .get_inner_tensor()?
+        .par_enum_map(|_, e| {
+            Ok::<ValType<F>, TensorError>(if let Some(f) = e.get_felt_eval() {
+                let col_idx = if !is_dummy {
+                    let table = config
+                        .range_checks
+                        .get(range)
+                        .ok_or(TensorError::TableLookupError)?;
+                    table.get_col_index(f)
+                } else {
+                    F::ZERO
+                };
+                Value::known(col_idx).into()
+            } else {
+                Value::<F>::unknown().into()
+            })
+        })?
+        .into();
+
+    region.assign(&config.lookup_index, &table_index)?;
+
     if !is_dummy {
         (0..assigned_len)
             .map(|i| {
                 let (x, y, z) = config
                     .lookup_input
                     .cartesian_coord(region.linear_coord() + i);
-                let selector = config.range_check_selectors.get(&(range.clone(), x, y));
+                let selector = config.range_check_selectors.get(&(*range, x, y));
                 region.enable(selector, z)?;
                 Ok(())
             })
             .collect::<Result<Vec<_>, Box<dyn Error>>>()?;
     }
 
+    if region.throw_range_check_error() {
+        // assert is within range
+        let int_values = w.get_int_evals()?;
+        for v in int_values {
+            if v < range.0 || v > range.1 {
+                log::debug!("Value ({:?}) out of range: {:?}", v, range);
+                return Err(Box::new(TensorError::TableLookupError));
+            }
+        }
+    }
+
     region.increment(assigned_len);
 
     let elapsed = timer.elapsed();
@@ -2498,7 +2684,7 @@ pub fn nonlinearity<F: PrimeField + TensorType + PartialOrd>(
     values: &[ValTensor<F>; 1],
     nl: &LookupOp,
 ) -> Result<ValTensor<F>, Box<dyn Error>> {
-    region.add_used_lookup(nl.clone());
+    region.add_used_lookup(nl.clone(), values)?;
 
     // time the entire operation
     let timer = instant::Instant::now();
@@ -2581,22 +2767,6 @@ 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>,
@@ -2709,24 +2879,8 @@ pub fn max<F: PrimeField + TensorType + PartialOrd>(
     )?;
     // relu(x - max(x - 1))
     let relu = nonlinearity(config, region, &[diff], &LookupOp::ReLU)?;
-
-    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);
+    // constraining relu(x - max(x - 1)) = 0/1
+    boolean_identity(config, region, &[relu.clone()], false)?;
 
     // sum(relu(x - max(x - 1)))
     let sum_relu = sum(config, region, &[relu])?;
@@ -2737,13 +2891,7 @@ 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
-    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());
+    is_zero_identity(config, region, &[relu_one_minus_sum_relu], false)?;
 
     Ok(assigned_max_val)
 }
@@ -2788,23 +2936,8 @@ pub fn min<F: PrimeField + TensorType + PartialOrd>(
 
     // relu(min(x + 1)  - x)
     let relu = nonlinearity(config, region, &[diff], &LookupOp::ReLU)?;
-
-    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);
+    // constraining relu(min(x + 1) - x) = 0/1
+    boolean_identity(config, region, &[relu.clone()], false)?;
 
     // sum(relu(min(x + 1) - x))
     let sum_relu = sum(config, region, &[relu])?;
@@ -2815,14 +2948,8 @@ 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
-    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());
+    is_zero_identity(config, region, &[relu_one_minus_sum_relu], false)?;
 
     Ok(assigned_min_val)
 }
@@ -2963,16 +3090,8 @@ pub fn softmax<F: PrimeField + TensorType + PartialOrd>(
     let denom = sum(config, region, &[ex.clone()])?;
     // get the inverse
 
-    let inv_denom = nonlinearity(
-        config,
-        region,
-        &[denom],
-        // we set to input scale + output_scale so the output scale is output)scale
-        &LookupOp::Recip {
-            input_scale: scale,
-            output_scale: scale,
-        },
-    )?;
+    let felt_scale = F::from(scale.0 as u64);
+    let inv_denom = recip(config, region, &[denom], felt_scale, felt_scale)?;
 
     // product of num * (1 / denom) = 2*output_scale
     let softmax = pairwise(config, region, &[ex, inv_denom], BaseOp::Mult)?;
@@ -2995,61 +3114,56 @@ pub fn range_check_percent<F: PrimeField + TensorType + PartialOrd>(
         return enforce_equality(config, region, values);
     }
 
-    // Calculate the difference between the expected output and actual output
-    let diff = pairwise(config, region, values, BaseOp::Sub)?;
+    let mut values = [values[0].clone(), values[1].clone()];
 
-    // Calculate the reciprocal of the expected output tensor, scaling by double the scaling factor
-    let recip = nonlinearity(
+    values[0] = region.assign(&config.inputs[0], &values[0])?;
+    values[1] = region.assign(&config.inputs[1], &values[1])?;
+    let total_assigned_0 = values[0].len();
+    let total_assigned_1 = values[1].len();
+    let total_assigned = std::cmp::max(total_assigned_0, total_assigned_1);
+    region.increment(total_assigned);
+
+    // Calculate the difference between the expected output and actual output
+    let diff = pairwise(config, region, &values, BaseOp::Sub)?;
+
+    // integer scale
+    let int_scale = scale.0 as i128;
+    // felt scale
+    let felt_scale = i128_to_felt(int_scale);
+    // range check len capped at 2^(S-3) and make it divisible 2
+    let range_check_bracket = std::cmp::min(
+        utils::F32(scale.0),
+        utils::F32(2_f32.powf((F::S - 5) as f32)),
+    )
+    .0;
+
+    let range_check_bracket_int = range_check_bracket as i128;
+
+    // input scale ratio we multiply by tol such that in the new scale range_check_len represents tol percent
+    let input_scale_ratio = ((scale.0.powf(2.0) / range_check_bracket) * tol) as i128 / 2 * 2;
+
+    let recip = recip(
         config,
         region,
         &[values[0].clone()],
-        &LookupOp::Recip {
-            input_scale: scale,
-            output_scale: scale,
-        },
+        felt_scale,
+        felt_scale * F::from(100),
     )?;
 
+    log::debug!("recip: {}", recip.show());
+
     // Multiply the difference by the recip
     let product = pairwise(config, region, &[diff, recip], BaseOp::Mult)?;
 
-    let scale_squared = scale.0 * scale.0;
+    log::debug!("product: {}", product.show());
+    let rebased_product = loop_div(config, region, &[product], i128_to_felt(input_scale_ratio))?;
+    log::debug!("rebased_product: {}", rebased_product.show());
 
-    // Use the greater than look up table to check if the percent error is within the tolerance for upper bound
-    let tol = tol / 100.0;
-    let upper_bound = nonlinearity(
+    // check that it is within the tolerance range
+    range_check(
         config,
         region,
-        &[product.clone()],
-        &LookupOp::GreaterThan {
-            a: utils::F32(tol * scale_squared),
-        },
-    )?;
-
-    // Negate the product
-    let neg_product = neg(config, region, &[product])?;
-
-    // Use the greater than look up table to check if the percent error is within the tolerance for lower bound
-    let lower_bound = nonlinearity(
-        config,
-        region,
-        &[neg_product],
-        &LookupOp::GreaterThan {
-            a: utils::F32(tol * scale_squared),
-        },
-    )?;
-
-    // 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
-    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)
+        &[rebased_product],
+        &(-range_check_bracket_int, range_check_bracket_int),
+    )
 }

src/circuit/ops/lookup.rs

@@ -227,10 +227,7 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for LookupOp {
 
         let output = res.map(|x| i128_to_felt(x));
 
-        Ok(ForwardResult {
-            output,
-            intermediate_lookups: vec![],
-        })
+        Ok(ForwardResult { output })
     }
 
     /// Returns the name of the operation
@@ -246,10 +243,10 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for LookupOp {
             LookupOp::Max { scale, a } => format!("MAX(scale={}, a={})", scale, a),
             LookupOp::Min { scale, a } => format!("MIN(scale={}, a={})", scale, a),
             LookupOp::Sign => "SIGN".into(),
-            LookupOp::GreaterThan { .. } => "GREATER_THAN".into(),
-            LookupOp::GreaterThanEqual { .. } => "GREATER_THAN_EQUAL".into(),
-            LookupOp::LessThan { .. } => "LESS_THAN".into(),
-            LookupOp::LessThanEqual { .. } => "LESS_THAN_EQUAL".into(),
+            LookupOp::GreaterThan { a } => format!("GREATER_THAN(a={})", a),
+            LookupOp::GreaterThanEqual { a } => format!("GREATER_THAN_EQUAL(a={})", a),
+            LookupOp::LessThan { a } => format!("LESS_THAN(a={})", a),
+            LookupOp::LessThanEqual { a } => format!("LESS_THAN_EQUAL(a={})", a),
             LookupOp::Recip {
                 input_scale,
                 output_scale,

src/circuit/ops/mod.rs

@@ -29,7 +29,6 @@ pub mod region;
 #[derive(Clone, Debug, PartialEq, Eq, PartialOrd, Ord, Serialize, Deserialize)]
 pub struct ForwardResult<F: PrimeField + TensorType + PartialOrd> {
     pub(crate) output: Tensor<F>,
-    pub(crate) intermediate_lookups: Vec<Tensor<i128>>,
 }
 
 /// A trait representing operations that can be represented as constraints in a circuit.
@@ -178,7 +177,6 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for Input {
     fn f(&self, x: &[Tensor<F>]) -> Result<ForwardResult<F>, TensorError> {
         Ok(ForwardResult {
             output: x[0].clone(),
-            intermediate_lookups: vec![],
         })
     }
 
@@ -201,6 +199,7 @@ impl<F: PrimeField + TensorType + PartialOrd> Op<F> for Input {
                         config,
                         region,
                         values[..].try_into()?,
+                        true,
                     )?))
                 }
                 _ => Ok(Some(super::layouts::identity(
@@ -303,10 +302,7 @@ impl<F: PrimeField + TensorType + PartialOrd + Serialize + for<'de> Deserialize<
     fn f(&self, _: &[Tensor<F>]) -> Result<ForwardResult<F>, TensorError> {
         let output = self.quantized_values.clone();
 
-        Ok(ForwardResult {
-            output,
-            intermediate_lookups: vec![],
-        })
+        Ok(ForwardResult { output })
     }
 
     fn as_string(&self) -> String {

src/circuit/ops/poly.rs

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

src/circuit/ops/region.rs

@@ -16,6 +16,8 @@ use std::{
     },
 };
 
+use portable_atomic::AtomicI128 as AtomicInt;
+
 use super::lookup::LookupOp;
 
 /// Region error
@@ -66,6 +68,10 @@ 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,
+    max_range_size: i128,
+    throw_range_check_error: bool,
 }
 
 impl<'a, F: PrimeField + TensorType + PartialOrd> RegionCtx<'a, F> {
@@ -74,6 +80,11 @@ impl<'a, F: PrimeField + TensorType + PartialOrd> RegionCtx<'a, F> {
         self.total_constants += n;
     }
 
+    ///
+    pub fn throw_range_check_error(&self) -> bool {
+        self.throw_range_check_error
+    }
+
     /// Create a new region context
     pub fn new(region: Region<'a, F>, row: usize, num_inner_cols: usize) -> RegionCtx<'a, F> {
         let region = Some(RefCell::new(region));
@@ -87,6 +98,10 @@ 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,
+            max_range_size: 0,
+            throw_range_check_error: false,
         }
     }
     /// Create a new region context from a wrapped region
@@ -104,11 +119,19 @@ 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,
+            max_range_size: 0,
+            throw_range_check_error: false,
         }
     }
 
     /// Create a new region context
-    pub fn new_dummy(row: usize, num_inner_cols: usize) -> RegionCtx<'a, F> {
+    pub fn new_dummy(
+        row: usize,
+        num_inner_cols: usize,
+        throw_range_check_error: bool,
+    ) -> RegionCtx<'a, F> {
         let region = None;
         let linear_coord = row * num_inner_cols;
 
@@ -120,6 +143,10 @@ 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,
+            max_range_size: 0,
+            throw_range_check_error,
         }
     }
 
@@ -131,6 +158,7 @@ impl<'a, F: PrimeField + TensorType + PartialOrd> RegionCtx<'a, F> {
         num_inner_cols: usize,
         used_lookups: HashSet<LookupOp>,
         used_range_checks: HashSet<Range>,
+        throw_range_check_error: bool,
     ) -> RegionCtx<'a, F> {
         let region = None;
         RegionCtx {
@@ -141,6 +169,10 @@ impl<'a, F: PrimeField + TensorType + PartialOrd> RegionCtx<'a, F> {
             total_constants,
             used_lookups,
             used_range_checks,
+            max_lookup_inputs: 0,
+            min_lookup_inputs: 0,
+            max_range_size: 0,
+            throw_range_check_error,
         }
     }
 
@@ -180,6 +212,7 @@ impl<'a, F: PrimeField + TensorType + PartialOrd> RegionCtx<'a, F> {
 
     /// Create a new region context per loop iteration
     /// hacky but it works
+
     pub fn dummy_loop<T: TensorType + Send + Sync>(
         &mut self,
         output: &mut Tensor<T>,
@@ -190,6 +223,8 @@ 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());
+        let min_lookup_inputs = AtomicInt::new(self.min_lookup_inputs());
         let lookups = Arc::new(Mutex::new(self.used_lookups.clone()));
         let range_checks = Arc::new(Mutex::new(self.used_range_checks.clone()));
 
@@ -209,6 +244,7 @@ impl<'a, F: PrimeField + TensorType + PartialOrd> RegionCtx<'a, F> {
                     self.num_inner_cols,
                     HashSet::new(),
                     HashSet::new(),
+                    self.throw_range_check_error,
                 );
                 let res = inner_loop_function(idx, &mut local_reg);
                 // we update the offset and constants
@@ -221,6 +257,9 @@ impl<'a, F: PrimeField + TensorType + PartialOrd> RegionCtx<'a, F> {
                     local_reg.total_constants() - starting_constants,
                     Ordering::SeqCst,
                 );
+
+                max_lookup_inputs.fetch_max(local_reg.max_lookup_inputs(), Ordering::SeqCst);
+                min_lookup_inputs.fetch_min(local_reg.min_lookup_inputs(), Ordering::SeqCst);
                 // update the lookups
                 let mut lookups = lookups.lock().unwrap();
                 lookups.extend(local_reg.used_lookups());
@@ -234,6 +273,11 @@ impl<'a, F: PrimeField + TensorType + PartialOrd> RegionCtx<'a, F> {
             })?;
         self.total_constants = constants.into_inner();
         self.linear_coord = linear_coord.into_inner();
+        #[allow(trivial_numeric_casts)]
+        {
+            self.max_lookup_inputs = max_lookup_inputs.into_inner();
+            self.min_lookup_inputs = min_lookup_inputs.into_inner();
+        }
         self.row = row.into_inner();
         self.used_lookups = Arc::try_unwrap(lookups)
             .map_err(|e| RegionError::from(format!("dummy_loop: failed to get lookups: {:?}", e)))?
@@ -253,19 +297,55 @@ 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());
+        }
+
+        let range_size = (range.1 - range.0).abs();
+
+        self.max_range_size = self.max_range_size.max(range_size);
+        Ok(())
+    }
+
     /// Check if the region is dummy
     pub fn is_dummy(&self) -> bool {
         self.region.is_none()
     }
 
     /// add used lookup
-    pub fn add_used_lookup(&mut self, lookup: LookupOp) {
+    pub fn add_used_lookup(
+        &mut self,
+        lookup: LookupOp,
+        inputs: &[ValTensor<F>],
+    ) -> Result<(), Box<dyn std::error::Error>> {
         self.used_lookups.insert(lookup);
+        self.update_max_min_lookup_inputs(inputs)
     }
 
     /// add used range check
-    pub fn add_used_range_check(&mut self, range: Range) {
+    pub fn add_used_range_check(&mut self, range: Range) -> Result<(), Box<dyn std::error::Error>> {
         self.used_range_checks.insert(range);
+        self.update_max_min_lookup_range(range)
     }
 
     /// Get the offset
@@ -293,6 +373,21 @@ 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
+    }
+
+    /// max range check
+    pub fn max_range_size(&self) -> i128 {
+        self.max_range_size
+    }
+
     /// Assign a constant value
     pub fn assign_constant(&mut self, var: &VarTensor, value: F) -> Result<ValType<F>, Error> {
         self.total_constants += 1;

src/circuit/table.rs

@@ -130,14 +130,12 @@ impl<F: PrimeField + TensorType + PartialOrd> Table<F> {
     pub fn cal_bit_range(bits: usize, reserved_blinding_rows: usize) -> usize {
         2usize.pow(bits as u32) - reserved_blinding_rows
     }
+}
 
-    ///
-    pub fn num_cols_required(range: Range, col_size: usize) -> usize {
-        // double it to be safe
-        let range_len = range.1 - range.0;
-        // number of cols needed to store the range
-        (range_len / (col_size as i128)) as usize + 1
-    }
+///
+pub fn num_cols_required(range_len: i128, col_size: usize) -> usize {
+    // number of cols needed to store the range
+    (range_len / (col_size as i128)) as usize + 1
 }
 
 impl<F: PrimeField + TensorType + PartialOrd> Table<F> {
@@ -152,7 +150,7 @@ impl<F: PrimeField + TensorType + PartialOrd> Table<F> {
         let factors = cs.blinding_factors() + RESERVED_BLINDING_ROWS_PAD;
         let col_size = Self::cal_col_size(logrows, factors);
         // number of cols needed to store the range
-        let num_cols = Self::num_cols_required(range, col_size);
+        let num_cols = num_cols_required((range.1 - range.0).abs(), col_size);
 
         log::debug!("table range: {:?}", range);
 
@@ -265,7 +263,9 @@ impl<F: PrimeField + TensorType + PartialOrd> Table<F> {
 #[derive(Clone, Debug)]
 pub struct RangeCheck<F: PrimeField> {
     /// Input to table.
-    pub input: TableColumn,
+    pub inputs: Vec<TableColumn>,
+    /// col size
+    pub col_size: usize,
     /// selector cn
     pub selector_constructor: SelectorConstructor<F>,
     /// Flags if table has been previously assigned to.
@@ -277,8 +277,10 @@ pub struct RangeCheck<F: PrimeField> {
 
 impl<F: PrimeField + TensorType + PartialOrd> RangeCheck<F> {
     /// get first_element of column
-    pub fn get_first_element(&self) -> F {
-        i128_to_felt(self.range.0)
+    pub fn get_first_element(&self, chunk: usize) -> F {
+        let chunk = chunk as i128;
+        // we index from 1 to prevent soundness issues
+        i128_to_felt(chunk * (self.col_size as i128) + self.range.0)
     }
 
     ///
@@ -290,24 +292,58 @@ impl<F: PrimeField + TensorType + PartialOrd> RangeCheck<F> {
     pub fn cal_bit_range(bits: usize, reserved_blinding_rows: usize) -> usize {
         2usize.pow(bits as u32) - reserved_blinding_rows
     }
+
+    /// get column index given input
+    pub fn get_col_index(&self, input: F) -> F {
+        //    range is split up into chunks of size col_size, find the chunk that input is in
+        let chunk =
+            (crate::fieldutils::felt_to_i128(input) - self.range.0).abs() / (self.col_size as i128);
+
+        i128_to_felt(chunk)
+    }
 }
 
 impl<F: PrimeField + TensorType + PartialOrd> RangeCheck<F> {
     /// Configures the table.
-    pub fn configure(cs: &mut ConstraintSystem<F>, range: Range) -> RangeCheck<F> {
+    pub fn configure(cs: &mut ConstraintSystem<F>, range: Range, logrows: usize) -> RangeCheck<F> {
         log::debug!("range check range: {:?}", range);
 
-        let inputs = cs.lookup_table_column();
+        let factors = cs.blinding_factors() + RESERVED_BLINDING_ROWS_PAD;
+        let col_size = Self::cal_col_size(logrows, factors);
+        // number of cols needed to store the range
+        let num_cols = num_cols_required((range.1 - range.0).abs(), col_size);
+
+        let inputs = {
+            let mut cols = vec![];
+            for _ in 0..num_cols {
+                cols.push(cs.lookup_table_column());
+            }
+            cols
+        };
+
+        let num_cols = inputs.len();
+
+        if num_cols > 1 {
+            warn!("Using {} columns for range-check.", num_cols);
+        }
 
         RangeCheck {
-            input: inputs,
+            inputs,
+            col_size,
             is_assigned: false,
-            selector_constructor: SelectorConstructor::new(2),
+            selector_constructor: SelectorConstructor::new(num_cols),
             range,
             _marker: PhantomData,
         }
     }
 
+    /// Take a linear coordinate and output the (column, row) position in the storage block.
+    pub fn cartesian_coord(&self, linear_coord: usize) -> (usize, usize) {
+        let x = linear_coord / self.col_size;
+        let y = linear_coord % self.col_size;
+        (x, y)
+    }
+
     /// Assigns values to the constraints generated when calling `configure`.
     pub fn layout(&mut self, layouter: &mut impl Layouter<F>) -> Result<(), Box<dyn Error>> {
         if self.is_assigned {
@@ -318,28 +354,43 @@ impl<F: PrimeField + TensorType + PartialOrd> RangeCheck<F> {
         let largest = self.range.1;
 
         let inputs: Tensor<F> = Tensor::from(smallest..=largest).map(|x| i128_to_felt(x));
+        let chunked_inputs = inputs.chunks(self.col_size);
 
         self.is_assigned = true;
 
-        layouter.assign_table(
-            || "range check table",
-            |mut table| {
-                let _ = inputs
-                    .iter()
-                    .enumerate()
-                    .map(|(row_offset, input)| {
-                        table.assign_cell(
-                            || format!("rc_i_col row {}", row_offset),
-                            self.input,
-                            row_offset,
-                            || Value::known(*input),
-                        )?;
+        let col_multipliers: Vec<F> = (0..chunked_inputs.len())
+            .map(|x| self.selector_constructor.get_selector_val_at_idx(x))
+            .collect();
+
+        let _ = chunked_inputs
+            .enumerate()
+            .map(|(chunk_idx, inputs)| {
+                layouter.assign_table(
+                    || "range check table",
+                    |mut table| {
+                        let _ = inputs
+                            .iter()
+                            .enumerate()
+                            .map(|(mut row_offset, input)| {
+                                let col_multiplier = col_multipliers[chunk_idx];
+
+                                row_offset += chunk_idx * self.col_size;
+                                let (x, y) = self.cartesian_coord(row_offset);
+                                table.assign_cell(
+                                    || format!("rc_i_col row {}", row_offset),
+                                    self.inputs[x],
+                                    y,
+                                    || Value::known(*input * col_multiplier),
+                                )?;
+
+                                Ok(())
+                            })
+                            .collect::<Result<Vec<()>, halo2_proofs::plonk::Error>>()?;
                         Ok(())
-                    })
-                    .collect::<Result<Vec<()>, halo2_proofs::plonk::Error>>()?;
-                Ok(())
-            },
-        )?;
+                    },
+                )
+            })
+            .collect::<Result<Vec<()>, halo2_proofs::plonk::Error>>()?;
         Ok(())
     }
 }

src/circuit/tests.rs

@@ -1,4 +1,3 @@
-use crate::circuit::ops::hybrid::HybridOp;
 use crate::circuit::ops::poly::PolyOp;
 use crate::circuit::*;
 use crate::tensor::{Tensor, TensorType, ValTensor, VarTensor};
@@ -246,7 +245,7 @@ mod matmul_col_overflow {
 #[cfg(test)]
 #[cfg(not(all(target_arch = "wasm32", target_os = "unknown")))]
 mod matmul_col_ultra_overflow_double_col {
-    use halo2_proofs::poly::commitment::ParamsProver;
+    use halo2_proofs::poly::commitment::{Params, ParamsProver};
 
     use super::*;
 
@@ -349,8 +348,13 @@ mod matmul_col_ultra_overflow_double_col {
         let strategy =
             halo2_proofs::poly::kzg::strategy::SingleStrategy::new(params.verifier_params());
         let vk = pk.get_vk();
-        let result =
-            crate::pfsys::verify_proof_circuit_kzg(params.verifier_params(), proof, vk, strategy);
+        let result = crate::pfsys::verify_proof_circuit_kzg(
+            params.verifier_params(),
+            proof,
+            vk,
+            strategy,
+            params.n(),
+        );
 
         assert!(result.is_ok());
 
@@ -361,7 +365,7 @@ mod matmul_col_ultra_overflow_double_col {
 #[cfg(test)]
 #[cfg(not(all(target_arch = "wasm32", target_os = "unknown")))]
 mod matmul_col_ultra_overflow {
-    use halo2_proofs::poly::commitment::ParamsProver;
+    use halo2_proofs::poly::commitment::{Params, ParamsProver};
 
     use super::*;
 
@@ -463,8 +467,13 @@ mod matmul_col_ultra_overflow {
         let strategy =
             halo2_proofs::poly::kzg::strategy::SingleStrategy::new(params.verifier_params());
         let vk = pk.get_vk();
-        let result =
-            crate::pfsys::verify_proof_circuit_kzg(params.verifier_params(), proof, vk, strategy);
+        let result = crate::pfsys::verify_proof_circuit_kzg(
+            params.verifier_params(),
+            proof,
+            vk,
+            strategy,
+            params.n(),
+        );
 
         assert!(result.is_ok());
 
@@ -1140,7 +1149,7 @@ mod conv {
 #[cfg(test)]
 #[cfg(not(all(target_arch = "wasm32", target_os = "unknown")))]
 mod conv_col_ultra_overflow {
-    use halo2_proofs::poly::commitment::ParamsProver;
+    use halo2_proofs::poly::commitment::{Params, ParamsProver};
 
     use super::*;
 
@@ -1262,8 +1271,13 @@ mod conv_col_ultra_overflow {
         let strategy =
             halo2_proofs::poly::kzg::strategy::SingleStrategy::new(params.verifier_params());
         let vk = pk.get_vk();
-        let result =
-            crate::pfsys::verify_proof_circuit_kzg(params.verifier_params(), proof, vk, strategy);
+        let result = crate::pfsys::verify_proof_circuit_kzg(
+            params.verifier_params(),
+            proof,
+            vk,
+            strategy,
+            params.n(),
+        );
 
         assert!(result.is_ok());
 
@@ -1275,7 +1289,7 @@ mod conv_col_ultra_overflow {
 // not wasm 32 unknown
 #[cfg(not(all(target_arch = "wasm32", target_os = "unknown")))]
 mod conv_relu_col_ultra_overflow {
-    use halo2_proofs::poly::commitment::ParamsProver;
+    use halo2_proofs::poly::commitment::{Params, ParamsProver};
 
     use super::*;
 
@@ -1412,8 +1426,13 @@ mod conv_relu_col_ultra_overflow {
         let strategy =
             halo2_proofs::poly::kzg::strategy::SingleStrategy::new(params.verifier_params());
         let vk = pk.get_vk();
-        let result =
-            crate::pfsys::verify_proof_circuit_kzg(params.verifier_params(), proof, vk, strategy);
+        let result = crate::pfsys::verify_proof_circuit_kzg(
+            params.verifier_params(),
+            proof,
+            vk,
+            strategy,
+            params.n(),
+        );
 
         assert!(result.is_ok());
 
@@ -2120,148 +2139,6 @@ mod matmul_relu {
     }
 }
 
-#[cfg(test)]
-mod rangecheckpercent {
-    use crate::circuit::Tolerance;
-    use crate::{circuit, tensor::Tensor};
-    use halo2_proofs::{
-        circuit::{Layouter, SimpleFloorPlanner, Value},
-        dev::MockProver,
-        plonk::{Circuit, ConstraintSystem, Error},
-    };
-
-    const RANGE: f32 = 1.0; // 1 percent error tolerance
-    const K: usize = 18;
-    const LEN: usize = 1;
-    const SCALE: usize = i128::pow(2, 7) as usize;
-
-    use super::*;
-
-    #[derive(Clone)]
-    struct MyCircuit<F: PrimeField + TensorType + PartialOrd> {
-        input: ValTensor<F>,
-        output: ValTensor<F>,
-        _marker: PhantomData<F>,
-    }
-
-    impl Circuit<F> for MyCircuit<F> {
-        type Config = BaseConfig<F>;
-        type FloorPlanner = SimpleFloorPlanner;
-        type Params = TestParams;
-
-        fn without_witnesses(&self) -> Self {
-            self.clone()
-        }
-
-        fn configure(cs: &mut ConstraintSystem<F>) -> Self::Config {
-            let scale = utils::F32(SCALE as f32);
-            let a = VarTensor::new_advice(cs, K, 1, LEN);
-            let b = VarTensor::new_advice(cs, K, 1, LEN);
-            let output = VarTensor::new_advice(cs, K, 1, LEN);
-            let mut config =
-                Self::Config::configure(cs, &[a.clone(), b.clone()], &output, CheckMode::SAFE);
-            // set up a new GreaterThan and Recip tables
-            let nl = &LookupOp::GreaterThan {
-                a: circuit::utils::F32((RANGE * SCALE.pow(2) as f32) / 100.0),
-            };
-            config
-                .configure_lookup(cs, &b, &output, &a, (-32768, 32768), K, nl)
-                .unwrap();
-
-            config
-                .configure_lookup(
-                    cs,
-                    &b,
-                    &output,
-                    &a,
-                    (-32768, 32768),
-                    K,
-                    &LookupOp::Recip {
-                        input_scale: scale,
-                        output_scale: scale,
-                    },
-                )
-                .unwrap();
-            config
-        }
-
-        fn synthesize(
-            &self,
-            mut config: Self::Config,
-            mut layouter: impl Layouter<F>,
-        ) -> Result<(), Error> {
-            config.layout_tables(&mut layouter).unwrap();
-            layouter
-                .assign_region(
-                    || "",
-                    |region| {
-                        let mut region = RegionCtx::new(region, 0, 1);
-                        config
-                            .layout(
-                                &mut region,
-                                &[self.output.clone(), self.input.clone()],
-                                Box::new(HybridOp::RangeCheck(Tolerance {
-                                    val: RANGE,
-                                    scale: SCALE.into(),
-                                })),
-                            )
-                            .map_err(|_| Error::Synthesis)
-                    },
-                )
-                .unwrap();
-            Ok(())
-        }
-    }
-
-    #[test]
-    #[allow(clippy::assertions_on_constants)]
-    fn test_range_check_percent() {
-        // Successful cases
-        {
-            let inp = Tensor::new(Some(&[Value::<F>::known(F::from(100_u64))]), &[1]).unwrap();
-            let out = Tensor::new(Some(&[Value::<F>::known(F::from(101_u64))]), &[1]).unwrap();
-            let circuit = MyCircuit::<F> {
-                input: ValTensor::from(inp),
-                output: ValTensor::from(out),
-                _marker: PhantomData,
-            };
-            let prover = MockProver::run(K as u32, &circuit, vec![]).unwrap();
-            prover.assert_satisfied();
-        }
-        {
-            let inp = Tensor::new(Some(&[Value::<F>::known(F::from(200_u64))]), &[1]).unwrap();
-            let out = Tensor::new(Some(&[Value::<F>::known(F::from(199_u64))]), &[1]).unwrap();
-            let circuit = MyCircuit::<F> {
-                input: ValTensor::from(inp),
-                output: ValTensor::from(out),
-                _marker: PhantomData,
-            };
-            let prover = MockProver::run(K as u32, &circuit, vec![]).unwrap();
-            prover.assert_satisfied();
-        }
-
-        // Unsuccessful case
-        {
-            let inp = Tensor::new(Some(&[Value::<F>::known(F::from(100_u64))]), &[1]).unwrap();
-            let out = Tensor::new(Some(&[Value::<F>::known(F::from(102_u64))]), &[1]).unwrap();
-            let circuit = MyCircuit::<F> {
-                input: ValTensor::from(inp),
-                output: ValTensor::from(out),
-                _marker: PhantomData,
-            };
-            let prover = MockProver::run(K as u32, &circuit, vec![]).unwrap();
-            match prover.verify() {
-                Ok(_) => {
-                    assert!(false)
-                }
-                Err(_) => {
-                    assert!(true)
-                }
-            }
-        }
-    }
-}
-
 #[cfg(test)]
 mod relu {
     use super::*;
@@ -2343,7 +2220,7 @@ mod lookup_ultra_overflow {
     use halo2_proofs::{
         circuit::{Layouter, SimpleFloorPlanner, Value},
         plonk::{Circuit, ConstraintSystem, Error},
-        poly::commitment::ParamsProver,
+        poly::commitment::{Params, ParamsProver},
     };
 
     #[derive(Clone)]
@@ -2447,121 +2324,16 @@ mod lookup_ultra_overflow {
         let strategy =
             halo2_proofs::poly::kzg::strategy::SingleStrategy::new(params.verifier_params());
         let vk = pk.get_vk();
-        let result =
-            crate::pfsys::verify_proof_circuit_kzg(params.verifier_params(), proof, vk, strategy);
+        let result = crate::pfsys::verify_proof_circuit_kzg(
+            params.verifier_params(),
+            proof,
+            vk,
+            strategy,
+            params.n(),
+        );
 
         assert!(result.is_ok());
 
         println!("done.");
     }
 }
-
-#[cfg(test)]
-mod softmax {
-
-    use super::*;
-    use halo2_proofs::{
-        circuit::{Layouter, SimpleFloorPlanner, Value},
-        dev::MockProver,
-        plonk::{Circuit, ConstraintSystem, Error},
-    };
-
-    const K: usize = 18;
-    const LEN: usize = 3;
-    const SCALE: f32 = 128.0;
-
-    #[derive(Clone)]
-    struct SoftmaxCircuit<F: PrimeField + TensorType + PartialOrd> {
-        pub input: ValTensor<F>,
-        _marker: PhantomData<F>,
-    }
-
-    impl Circuit<F> for SoftmaxCircuit<F> {
-        type Config = BaseConfig<F>;
-        type FloorPlanner = SimpleFloorPlanner;
-        type Params = TestParams;
-
-        fn without_witnesses(&self) -> Self {
-            self.clone()
-        }
-        fn configure(cs: &mut ConstraintSystem<F>) -> Self::Config {
-            let a = VarTensor::new_advice(cs, K, 1, LEN);
-            let b = VarTensor::new_advice(cs, K, 1, LEN);
-            let output = VarTensor::new_advice(cs, K, 1, LEN);
-            let mut config = Self::Config::configure(cs, &[a, b], &output, CheckMode::SAFE);
-            let advices = (0..3)
-                .map(|_| VarTensor::new_advice(cs, K, 1, LEN))
-                .collect::<Vec<_>>();
-
-            config
-                .configure_lookup(
-                    cs,
-                    &advices[0],
-                    &advices[1],
-                    &advices[2],
-                    (-32768, 32768),
-                    K,
-                    &LookupOp::Exp {
-                        scale: SCALE.into(),
-                    },
-                )
-                .unwrap();
-            config
-                .configure_lookup(
-                    cs,
-                    &advices[0],
-                    &advices[1],
-                    &advices[2],
-                    (-32768, 32768),
-                    K,
-                    &LookupOp::Recip {
-                        input_scale: SCALE.into(),
-                        output_scale: SCALE.into(),
-                    },
-                )
-                .unwrap();
-            config
-        }
-
-        fn synthesize(
-            &self,
-            mut config: Self::Config,
-            mut layouter: impl Layouter<F>,
-        ) -> Result<(), Error> {
-            config.layout_tables(&mut layouter).unwrap();
-            layouter
-                .assign_region(
-                    || "",
-                    |region| {
-                        let mut region = RegionCtx::new(region, 0, 1);
-                        let _output = config
-                            .layout(
-                                &mut region,
-                                &[self.input.clone()],
-                                Box::new(HybridOp::Softmax {
-                                    scale: SCALE.into(),
-                                    axes: vec![0],
-                                }),
-                            )
-                            .unwrap();
-                        Ok(())
-                    },
-                )
-                .unwrap();
-
-            Ok(())
-        }
-    }
-
-    #[test]
-    fn softmax_circuit() {
-        let input = Tensor::from((0..LEN).map(|i| Value::known(F::from(i as u64 + 1))));
-
-        let circuit = SoftmaxCircuit::<F> {
-            input: ValTensor::from(input),
-            _marker: PhantomData,
-        };
-        let prover = MockProver::run(K as u32, &circuit, vec![]).unwrap();
-        prover.assert_satisfied();
-    }
-}

src/commands.rs

@@ -1,4 +1,4 @@
-use clap::{Parser, Subcommand, ValueEnum};
+use clap::{Parser, Subcommand};
 #[cfg(not(target_arch = "wasm32"))]
 use ethers::types::H160;
 #[cfg(feature = "python-bindings")]
@@ -9,8 +9,9 @@ use pyo3::{
     types::PyString,
 };
 use serde::{Deserialize, Serialize};
-use std::error::Error;
 use std::path::PathBuf;
+use std::{error::Error, str::FromStr};
+use tosubcommand::{ToFlags, ToSubcommand};
 
 use crate::{pfsys::ProofType, RunArgs};
 
@@ -85,15 +86,11 @@ pub const DEFAULT_VK_SOL: &str = "vk.sol";
 pub const DEFAULT_VK_ABI: &str = "vk.abi";
 /// Default scale rebase multipliers for calibration
 pub const DEFAULT_SCALE_REBASE_MULTIPLIERS: &str = "1,2,10";
+/// Default use reduced srs for verification
+pub const DEFAULT_USE_REDUCED_SRS_FOR_VERIFICATION: &str = "false";
+/// Default only check for range check rebase
+pub const DEFAULT_ONLY_RANGE_CHECK_REBASE: &str = "false";
 
-impl std::fmt::Display for TranscriptType {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        self.to_possible_value()
-            .expect("no values are skipped")
-            .get_name()
-            .fmt(f)
-    }
-}
 #[cfg(feature = "python-bindings")]
 /// Converts TranscriptType into a PyObject (Required for TranscriptType to be compatible with Python)
 impl IntoPy<PyObject> for TranscriptType {
@@ -138,17 +135,27 @@ impl Default for CalibrationTarget {
     }
 }
 
-impl ToString for CalibrationTarget {
-    fn to_string(&self) -> String {
-        match self {
-            CalibrationTarget::Resources { col_overflow: true } => {
-                "resources/col-overflow".to_string()
+impl std::fmt::Display for CalibrationTarget {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(
+            f,
+            "{}",
+            match self {
+                CalibrationTarget::Resources { col_overflow: true } => {
+                    "resources/col-overflow".to_string()
+                }
+                CalibrationTarget::Resources {
+                    col_overflow: false,
+                } => "resources".to_string(),
+                CalibrationTarget::Accuracy => "accuracy".to_string(),
             }
-            CalibrationTarget::Resources {
-                col_overflow: false,
-            } => "resources".to_string(),
-            CalibrationTarget::Accuracy => "accuracy".to_string(),
-        }
+        )
+    }
+}
+
+impl ToFlags for CalibrationTarget {
+    fn to_flags(&self) -> Vec<String> {
+        vec![format!("{}", self)]
     }
 }
 
@@ -169,6 +176,36 @@ impl From<&str> for CalibrationTarget {
     }
 }
 
+#[cfg(not(target_arch = "wasm32"))]
+#[derive(Debug, Copy, Clone, Serialize, Deserialize, PartialEq, PartialOrd)]
+/// wrapper for H160 to make it easy to parse into flag vals
+pub struct H160Flag {
+    inner: H160,
+}
+
+#[cfg(not(target_arch = "wasm32"))]
+impl From<H160Flag> for H160 {
+    fn from(val: H160Flag) -> H160 {
+        val.inner
+    }
+}
+
+#[cfg(not(target_arch = "wasm32"))]
+impl ToFlags for H160Flag {
+    fn to_flags(&self) -> Vec<String> {
+        vec![format!("{:#x}", self.inner)]
+    }
+}
+
+#[cfg(not(target_arch = "wasm32"))]
+impl From<&str> for H160Flag {
+    fn from(s: &str) -> Self {
+        Self {
+            inner: H160::from_str(s).unwrap(),
+        }
+    }
+}
+
 #[cfg(feature = "python-bindings")]
 /// Converts CalibrationTarget into a PyObject (Required for CalibrationTarget to be compatible with Python)
 impl IntoPy<PyObject> for CalibrationTarget {
@@ -201,7 +238,7 @@ impl<'source> FromPyObject<'source> for CalibrationTarget {
         }
     }
 }
-
+// not wasm
 use lazy_static::lazy_static;
 
 // if CARGO VERSION is 0.0.0 replace with "source - no compatibility guaranteed"
@@ -242,7 +279,7 @@ impl Cli {
 }
 
 #[allow(missing_docs)]
-#[derive(Debug, Subcommand, Clone, Deserialize, Serialize, PartialEq, PartialOrd)]
+#[derive(Debug, Subcommand, Clone, Deserialize, Serialize, PartialEq, PartialOrd, ToSubcommand)]
 pub enum Commands {
     #[cfg(feature = "empty-cmd")]
     /// Creates an empty buffer
@@ -336,9 +373,9 @@ pub enum Commands {
         /// max logrows to use for calibration, 26 is the max public SRS size
         #[arg(long)]
         max_logrows: Option<u32>,
-        // whether to fix the div_rebasing value truthiness during calibration. this changes how we rebase
-        #[arg(long)]
-        div_rebasing: Option<bool>,
+        // whether to only range check rebases (instead of trying both range check and lookup)
+        #[arg(long, default_value = DEFAULT_ONLY_RANGE_CHECK_REBASE)]
+        only_range_check_rebase: bool,
     },
 
     /// Generates a dummy SRS
@@ -434,7 +471,7 @@ pub enum Commands {
             long,
             require_equals = true,
             num_args = 0..=1,
-            default_value_t = TranscriptType::EVM,
+            default_value_t = TranscriptType::default(),
             value_enum
         )]
         transcript: TranscriptType,
@@ -489,13 +526,13 @@ pub enum Commands {
         #[arg(short = 'W', long, default_value = DEFAULT_WITNESS)]
         witness: PathBuf,
         /// The path to the compiled model file (generated using the compile-circuit command)
-        #[arg(short = 'M', long, default_value = DEFAULT_COMPILED_CIRCUIT)]
+        #[arg(short = 'M', long)]
         compiled_circuit: PathBuf,
         #[arg(
             long,
             require_equals = true,
             num_args = 0..=1,
-            default_value_t = TranscriptType::EVM,
+            default_value_t = TranscriptType::default(),
             value_enum
         )]
         transcript: TranscriptType,
@@ -509,7 +546,7 @@ pub enum Commands {
     #[cfg(not(target_arch = "wasm32"))]
     /// Deploys a test contact that the data attester reads from and creates a data attestation formatted input.json file that contains call data information
     #[command(arg_required_else_help = true)]
-    SetupTestEVMData {
+    SetupTestEvmData {
         /// The path to the .json data file, which should include both the network input (possibly private) and the network output (public input to the proof)
         #[arg(short = 'D', long)]
         data: PathBuf,
@@ -537,7 +574,7 @@ pub enum Commands {
     TestUpdateAccountCalls {
         /// The path to the verifier contract's address
         #[arg(long)]
-        addr: H160,
+        addr: H160Flag,
         /// The path to the .json data file.
         #[arg(short = 'D', long)]
         data: PathBuf,
@@ -587,9 +624,9 @@ pub enum Commands {
         check_mode: CheckMode,
     },
     #[cfg(not(target_arch = "wasm32"))]
-    /// Creates an EVM verifier for a single proof
+    /// Creates an Evm verifier for a single proof
     #[command(name = "create-evm-verifier")]
-    CreateEVMVerifier {
+    CreateEvmVerifier {
         /// The path to SRS, if None will use $EZKL_REPO_PATH/srs/kzg{logrows}.srs
         #[arg(long)]
         srs_path: Option<PathBuf>,
@@ -612,9 +649,9 @@ pub enum Commands {
         render_vk_seperately: bool,
     },
     #[cfg(not(target_arch = "wasm32"))]
-    /// Creates an EVM verifier for a single proof
+    /// Creates an Evm verifier for a single proof
     #[command(name = "create-evm-vk")]
-    CreateEVMVK {
+    CreateEvmVK {
         /// The path to SRS, if None will use $EZKL_REPO_PATH/srs/kzg{logrows}.srs
         #[arg(long)]
         srs_path: Option<PathBuf>,
@@ -632,9 +669,9 @@ pub enum Commands {
         abi_path: PathBuf,
     },
     #[cfg(not(target_arch = "wasm32"))]
-    /// Creates an EVM verifier that attests to on-chain inputs for a single proof
+    /// Creates an Evm verifier that attests to on-chain inputs for a single proof
     #[command(name = "create-evm-da")]
-    CreateEVMDataAttestation {
+    CreateEvmDataAttestation {
         /// The path to load circuit settings .json file from (generated using the gen-settings command)
         #[arg(short = 'S', long, default_value = DEFAULT_SETTINGS)]
         settings_path: PathBuf,
@@ -654,9 +691,9 @@ pub enum Commands {
     },
 
     #[cfg(not(target_arch = "wasm32"))]
-    /// Creates an EVM verifier for an aggregate proof
+    /// Creates an Evm verifier for an aggregate proof
     #[command(name = "create-evm-verifier-aggr")]
-    CreateEVMVerifierAggr {
+    CreateEvmVerifierAggr {
         /// The path to SRS, if None will use $EZKL_REPO_PATH/srs/kzg{logrows}.srs
         #[arg(long)]
         srs_path: Option<PathBuf>,
@@ -695,6 +732,9 @@ pub enum Commands {
         /// The path to SRS, if None will use $EZKL_REPO_PATH/srs/kzg{logrows}.srs
         #[arg(long)]
         srs_path: Option<PathBuf>,
+        /// Reduce SRS logrows to the number of instances rather than the number of logrows used for proofs (only works if the srs were generated in the same ceremony)
+        #[arg(long, default_value = DEFAULT_USE_REDUCED_SRS_FOR_VERIFICATION)]
+        reduced_srs: bool,
     },
     /// Verifies an aggregate proof, returning accept or reject
     VerifyAggr {
@@ -776,31 +816,23 @@ pub enum Commands {
         private_key: Option<String>,
     },
     #[cfg(not(target_arch = "wasm32"))]
-    /// Verifies a proof using a local EVM executor, returning accept or reject
+    /// Verifies a proof using a local Evm executor, returning accept or reject
     #[command(name = "verify-evm")]
-    VerifyEVM {
+    VerifyEvm {
         /// The path to the proof file (generated using the prove command)
         #[arg(long, default_value = DEFAULT_PROOF)]
         proof_path: PathBuf,
         /// The path to verifier contract's address
         #[arg(long, default_value = DEFAULT_CONTRACT_ADDRESS)]
-        addr_verifier: H160,
+        addr_verifier: H160Flag,
         /// RPC URL for an Ethereum node, if None will use Anvil but WON'T persist state
         #[arg(short = 'U', long)]
         rpc_url: Option<String>,
         /// does the verifier use data attestation ?
         #[arg(long)]
-        addr_da: Option<H160>,
+        addr_da: Option<H160Flag>,
         // is the vk rendered seperately, if so specify an address
         #[arg(long)]
-        addr_vk: Option<H160>,
-    },
-
-    /// Print the proof in hexadecimal
-    #[command(name = "print-proof-hex")]
-    PrintProofHex {
-        /// The path to the proof file
-        #[arg(long, default_value = DEFAULT_PROOF)]
-        proof_path: PathBuf,
+        addr_vk: Option<H160Flag>,
     },
 }

src/execute.rs

@@ -3,6 +3,8 @@ use crate::circuit::CheckMode;
 use crate::commands::CalibrationTarget;
 use crate::commands::Commands;
 #[cfg(not(target_arch = "wasm32"))]
+use crate::commands::H160Flag;
+#[cfg(not(target_arch = "wasm32"))]
 use crate::eth::{deploy_contract_via_solidity, deploy_da_verifier_via_solidity};
 #[cfg(not(target_arch = "wasm32"))]
 #[allow(unused_imports)]
@@ -21,8 +23,6 @@ use crate::pfsys::{create_proof_circuit_kzg, verify_proof_circuit_kzg};
 use crate::pfsys::{save_vk, srs::*};
 use crate::tensor::TensorError;
 use crate::RunArgs;
-#[cfg(not(target_arch = "wasm32"))]
-use ethers::types::H160;
 use gag::Gag;
 use halo2_proofs::dev::VerifyFailure;
 use halo2_proofs::poly::commitment::Params;
@@ -178,7 +178,7 @@ pub async fn run(command: Commands) -> Result<String, Box<dyn Error>> {
             scales,
             scale_rebase_multiplier,
             max_logrows,
-            div_rebasing,
+            only_range_check_rebase,
         } => calibrate(
             model,
             data,
@@ -187,7 +187,7 @@ pub async fn run(command: Commands) -> Result<String, Box<dyn Error>> {
             lookup_safety_margin,
             scales,
             scale_rebase_multiplier,
-            div_rebasing,
+            only_range_check_rebase,
             max_logrows,
         )
         .map(|e| serde_json::to_string(&e).unwrap()),
@@ -202,7 +202,7 @@ pub async fn run(command: Commands) -> Result<String, Box<dyn Error>> {
             .map(|e| serde_json::to_string(&e).unwrap()),
         Commands::Mock { model, witness } => mock(model, witness),
         #[cfg(not(target_arch = "wasm32"))]
-        Commands::CreateEVMVerifier {
+        Commands::CreateEvmVerifier {
             vk_path,
             srs_path,
             settings_path,
@@ -217,7 +217,7 @@ pub async fn run(command: Commands) -> Result<String, Box<dyn Error>> {
             abi_path,
             render_vk_seperately,
         ),
-        Commands::CreateEVMVK {
+        Commands::CreateEvmVK {
             vk_path,
             srs_path,
             settings_path,
@@ -225,14 +225,14 @@ pub async fn run(command: Commands) -> Result<String, Box<dyn Error>> {
             abi_path,
         } => create_evm_vk(vk_path, srs_path, settings_path, sol_code_path, abi_path),
         #[cfg(not(target_arch = "wasm32"))]
-        Commands::CreateEVMDataAttestation {
+        Commands::CreateEvmDataAttestation {
             settings_path,
             sol_code_path,
             abi_path,
             data,
         } => create_evm_data_attestation(settings_path, sol_code_path, abi_path, data),
         #[cfg(not(target_arch = "wasm32"))]
-        Commands::CreateEVMVerifierAggr {
+        Commands::CreateEvmVerifierAggr {
             vk_path,
             srs_path,
             sol_code_path,
@@ -270,7 +270,7 @@ pub async fn run(command: Commands) -> Result<String, Box<dyn Error>> {
             compress_selectors,
         ),
         #[cfg(not(target_arch = "wasm32"))]
-        Commands::SetupTestEVMData {
+        Commands::SetupTestEvmData {
             data,
             compiled_circuit,
             test_data,
@@ -366,7 +366,8 @@ pub async fn run(command: Commands) -> Result<String, Box<dyn Error>> {
             settings_path,
             vk_path,
             srs_path,
-        } => verify(proof_path, settings_path, vk_path, srs_path)
+            reduced_srs,
+        } => verify(proof_path, settings_path, vk_path, srs_path, reduced_srs)
             .map(|e| serde_json::to_string(&e).unwrap()),
         Commands::VerifyAggr {
             proof_path,
@@ -433,14 +434,13 @@ pub async fn run(command: Commands) -> Result<String, Box<dyn Error>> {
             .await
         }
         #[cfg(not(target_arch = "wasm32"))]
-        Commands::VerifyEVM {
+        Commands::VerifyEvm {
             proof_path,
             addr_verifier,
             rpc_url,
             addr_da,
             addr_vk,
         } => verify_evm(proof_path, addr_verifier, rpc_url, addr_da, addr_vk).await,
-        Commands::PrintProofHex { proof_path } => print_proof_hex(proof_path),
     }
 }
 
@@ -488,8 +488,14 @@ async fn fetch_srs(uri: &str) -> Result<Vec<u8>, Box<dyn Error>> {
 
 #[cfg(not(target_arch = "wasm32"))]
 fn check_srs_hash(logrows: u32, srs_path: Option<PathBuf>) -> Result<String, Box<dyn Error>> {
+    use std::io::Read;
+
     let path = get_srs_path(logrows, srs_path);
-    let hash = sha256::digest(std::fs::read(path.clone())?);
+    let file = std::fs::File::open(path.clone())?;
+    let mut buffer = vec![];
+    let bytes_read = std::io::BufReader::new(file).read_to_end(&mut buffer)?;
+    debug!("read {} bytes from SRS file", bytes_read);
+    let hash = sha256::digest(buffer);
     info!("SRS hash: {}", hash);
 
     let predefined_hash = match { crate::srs_sha::PUBLIC_SRS_SHA256_HASHES.get(&logrows) } {
@@ -612,7 +618,7 @@ pub(crate) async fn gen_witness(
 
     let start_time = Instant::now();
 
-    let witness = circuit.forward(&mut input, vk.as_ref(), srs.as_ref())?;
+    let witness = circuit.forward(&mut input, vk.as_ref(), srs.as_ref(), false)?;
 
     // print each variable tuple (symbol, value) as symbol=value
     trace!(
@@ -628,7 +634,7 @@ pub(crate) async fn gen_witness(
     );
 
     if let Some(output_path) = output {
-        serde_json::to_writer(&File::create(output_path)?, &witness)?;
+        witness.save(output_path)?;
     }
 
     // print the witness in debug
@@ -726,11 +732,11 @@ 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.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());
+            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);
         }
 
         let mean_percent_error =
@@ -780,6 +786,7 @@ impl AccuracyResults {
 /// Calibrate the circuit parameters to a given a dataset
 #[cfg(not(target_arch = "wasm32"))]
 #[allow(trivial_casts)]
+#[allow(clippy::too_many_arguments)]
 pub(crate) fn calibrate(
     model_path: PathBuf,
     data: PathBuf,
@@ -788,7 +795,7 @@ pub(crate) fn calibrate(
     lookup_safety_margin: i128,
     scales: Option<Vec<crate::Scale>>,
     scale_rebase_multiplier: Vec<u32>,
-    div_rebasing: Option<bool>,
+    only_range_check_rebase: bool,
     max_logrows: Option<u32>,
 ) -> Result<GraphSettings, Box<dyn Error>> {
     use std::collections::HashMap;
@@ -801,16 +808,7 @@ pub(crate) fn calibrate(
     // we load the model to get the input and output shapes
     // check if gag already exists
 
-    #[cfg(unix)]
-    let _r = match Gag::stdout() {
-        Ok(r) => Some(r),
-        Err(_) => None,
-    };
-
     let model = Model::from_run_args(&settings.run_args, &model_path)?;
-    // drop the gag
-    #[cfg(unix)]
-    std::mem::drop(_r);
 
     let chunks = data.split_into_batches(model.graph.input_shapes()?)?;
     info!("num of calibration batches: {}", chunks.len());
@@ -826,14 +824,11 @@ pub(crate) fn calibrate(
     let range = if let Some(scales) = scales {
         scales
     } else {
-        match target {
-            CalibrationTarget::Resources { .. } => (8..10).collect::<Vec<crate::Scale>>(),
-            CalibrationTarget::Accuracy => (10..14).collect::<Vec<crate::Scale>>(),
-        }
+        (11..14).collect::<Vec<crate::Scale>>()
     };
 
-    let div_rebasing = if let Some(div_rebasing) = div_rebasing {
-        vec![div_rebasing]
+    let div_rebasing = if only_range_check_rebase {
+        vec![false]
     } else {
         vec![true, false]
     };
@@ -892,16 +887,6 @@ pub(crate) fn calibrate(
             input_scale, param_scale, scale_rebase_multiplier, div_rebasing
         ));
 
-        #[cfg(unix)]
-        let _r = match Gag::stdout() {
-            Ok(r) => Some(r),
-            Err(_) => None,
-        };
-        #[cfg(unix)]
-        let _q = match Gag::stderr() {
-            Ok(r) => Some(r),
-            Err(_) => None,
-        };
         let key = (input_scale, param_scale, scale_rebase_multiplier);
         forward_pass_res.insert(key, vec![]);
 
@@ -916,17 +901,12 @@ pub(crate) fn calibrate(
         let mut circuit = match GraphCircuit::from_run_args(&local_run_args, &model_path) {
             Ok(c) => c,
             Err(e) => {
-                // drop the gag
-                #[cfg(unix)]
-                std::mem::drop(_r);
-                #[cfg(unix)]
-                std::mem::drop(_q);
                 debug!("circuit creation from run args failed: {:?}", e);
                 continue;
             }
         };
 
-        chunks
+        let forward_res = chunks
             .iter()
             .map(|chunk| {
                 let chunk = chunk.clone();
@@ -936,7 +916,7 @@ pub(crate) fn calibrate(
                     .map_err(|e| format!("failed to load circuit inputs: {}", e))?;
 
                 let forward_res = circuit
-                    .forward(&mut data.clone(), None, None)
+                    .forward(&mut data.clone(), None, None, true)
                     .map_err(|e| format!("failed to forward: {}", e))?;
 
                 // push result to the hashmap
@@ -947,7 +927,16 @@ pub(crate) fn calibrate(
 
                 Ok(()) as Result<(), String>
             })
-            .collect::<Result<Vec<()>, String>>()?;
+            .collect::<Result<Vec<()>, String>>();
+
+        match forward_res {
+            Ok(_) => (),
+            // typically errors will be due to the circuit overflowing the i128 limit
+            Err(e) => {
+                debug!("forward pass failed: {:?}", e);
+                continue;
+            }
+        }
 
         let min_lookup_range = forward_pass_res
             .get(&key)
@@ -965,19 +954,21 @@ pub(crate) fn calibrate(
             .max()
             .unwrap_or(0);
 
+        let max_range_size = forward_pass_res
+            .get(&key)
+            .unwrap()
+            .iter()
+            .map(|x| x.max_range_size)
+            .max()
+            .unwrap_or(0);
+
         let res = circuit.calibrate_from_min_max(
-            min_lookup_range,
-            max_lookup_range,
+            (min_lookup_range, max_lookup_range),
+            max_range_size,
             max_logrows,
             lookup_safety_margin,
         );
 
-        // drop the gag
-        #[cfg(unix)]
-        std::mem::drop(_r);
-        #[cfg(unix)]
-        std::mem::drop(_q);
-
         if res.is_ok() {
             let new_settings = circuit.settings().clone();
 
@@ -1172,16 +1163,6 @@ pub(crate) fn mock(
     Ok(String::new())
 }
 
-pub(crate) fn print_proof_hex(proof_path: PathBuf) -> Result<String, Box<dyn Error>> {
-    let proof = Snark::load::<KZGCommitmentScheme<Bn256>>(&proof_path)?;
-    for instance in proof.instances {
-        println!("{:?}", instance);
-    }
-    let hex_str = hex::encode(proof.proof);
-    info!("0x{}", hex_str);
-    Ok(format!("0x{}", hex_str))
-}
-
 #[cfg(feature = "render")]
 pub(crate) fn render(
     model: PathBuf,
@@ -1400,10 +1381,10 @@ pub(crate) async fn deploy_evm(
 #[cfg(not(target_arch = "wasm32"))]
 pub(crate) async fn verify_evm(
     proof_path: PathBuf,
-    addr_verifier: H160,
+    addr_verifier: H160Flag,
     rpc_url: Option<String>,
-    addr_da: Option<H160>,
-    addr_vk: Option<H160>,
+    addr_da: Option<H160Flag>,
+    addr_vk: Option<H160Flag>,
 ) -> Result<String, Box<dyn Error>> {
     use crate::eth::verify_proof_with_data_attestation;
     check_solc_requirement();
@@ -1413,14 +1394,20 @@ pub(crate) async fn verify_evm(
     let result = if let Some(addr_da) = addr_da {
         verify_proof_with_data_attestation(
             proof.clone(),
-            addr_verifier,
-            addr_da,
-            addr_vk,
+            addr_verifier.into(),
+            addr_da.into(),
+            addr_vk.map(|s| s.into()),
             rpc_url.as_deref(),
         )
         .await?
     } else {
-        verify_proof_via_solidity(proof.clone(), addr_verifier, addr_vk, rpc_url.as_deref()).await?
+        verify_proof_via_solidity(
+            proof.clone(),
+            addr_verifier.into(),
+            addr_vk.map(|s| s.into()),
+            rpc_url.as_deref(),
+        )
+        .await?
     };
 
     info!("Solidity verification result: {}", result);
@@ -1576,14 +1563,14 @@ pub(crate) async fn setup_test_evm_witness(
 use crate::pfsys::ProofType;
 #[cfg(not(target_arch = "wasm32"))]
 pub(crate) async fn test_update_account_calls(
-    addr: H160,
+    addr: H160Flag,
     data: PathBuf,
     rpc_url: Option<String>,
 ) -> Result<String, Box<dyn Error>> {
     use crate::eth::update_account_calls;
 
     check_solc_requirement();
-    update_account_calls(addr, data, rpc_url.as_deref()).await?;
+    update_account_calls(addr.into(), data, rpc_url.as_deref()).await?;
 
     Ok(String::new())
 }
@@ -1728,6 +1715,7 @@ pub(crate) fn fuzz(
             bad_proof,
             pk.get_vk(),
             strategy.clone(),
+            params.n(),
         )
         .map_err(|_| ())
     };
@@ -1758,6 +1746,7 @@ pub(crate) fn fuzz(
             bad_proof,
             pk.get_vk(),
             strategy.clone(),
+            params.n(),
         )
         .map_err(|_| ())
     };
@@ -1794,6 +1783,7 @@ pub(crate) fn fuzz(
             proof.clone(),
             bad_vk,
             strategy.clone(),
+            params.n(),
         )
         .map_err(|_| ())
     };
@@ -1825,6 +1815,7 @@ pub(crate) fn fuzz(
             bad_proof,
             pk.get_vk(),
             strategy.clone(),
+            params.n(),
         )
         .map_err(|_| ())
     };
@@ -1860,6 +1851,7 @@ pub(crate) fn fuzz(
             bad_proof,
             pk.get_vk(),
             strategy.clone(),
+            params.n(),
         )
         .map_err(|_| ())
     };
@@ -2045,15 +2037,29 @@ pub(crate) fn verify(
     settings_path: PathBuf,
     vk_path: PathBuf,
     srs_path: Option<PathBuf>,
+    reduced_srs: bool,
 ) -> Result<bool, Box<dyn Error>> {
     let circuit_settings = GraphSettings::load(&settings_path)?;
-    let params = load_params_cmd(srs_path, circuit_settings.run_args.logrows)?;
+
+    let params = if reduced_srs {
+        load_params_cmd(srs_path, circuit_settings.log2_total_instances())?
+    } else {
+        load_params_cmd(srs_path, circuit_settings.run_args.logrows)?
+    };
+
     let proof = Snark::load::<KZGCommitmentScheme<Bn256>>(&proof_path)?;
 
     let strategy = KZGSingleStrategy::new(params.verifier_params());
-    let vk = load_vk::<KZGCommitmentScheme<Bn256>, Fr, GraphCircuit>(vk_path, circuit_settings)?;
+    let vk =
+        load_vk::<KZGCommitmentScheme<Bn256>, Fr, GraphCircuit>(vk_path, circuit_settings.clone())?;
     let now = Instant::now();
-    let result = verify_proof_circuit_kzg(params.verifier_params(), proof, &vk, strategy);
+    let result = verify_proof_circuit_kzg(
+        params.verifier_params(),
+        proof,
+        &vk,
+        strategy,
+        1 << circuit_settings.run_args.logrows,
+    );
     let elapsed = now.elapsed();
     info!(
         "verify took {}.{}",
@@ -2077,7 +2083,7 @@ pub(crate) fn verify_aggr(
     let strategy = AccumulatorStrategy::new(params.verifier_params());
     let vk = load_vk::<KZGCommitmentScheme<Bn256>, Fr, AggregationCircuit>(vk_path, ())?;
     let now = Instant::now();
-    let result = verify_proof_circuit_kzg(&params, proof, &vk, strategy);
+    let result = verify_proof_circuit_kzg(&params, proof, &vk, strategy, 1 << logrows);
 
     let elapsed = now.elapsed();
     info!(

src/graph/input.rs

@@ -4,6 +4,7 @@ use crate::circuit::InputType;
 use crate::fieldutils::i128_to_felt;
 #[cfg(not(target_arch = "wasm32"))]
 use crate::tensor::Tensor;
+use crate::EZKL_BUF_CAPACITY;
 use halo2curves::bn256::Fr as Fp;
 #[cfg(not(target_arch = "wasm32"))]
 use postgres::{Client, NoTls};
@@ -15,6 +16,8 @@ use pyo3::types::PyDict;
 use pyo3::ToPyObject;
 use serde::ser::SerializeStruct;
 use serde::{Deserialize, Deserializer, Serialize, Serializer};
+use std::io::BufReader;
+use std::io::BufWriter;
 use std::io::Read;
 use std::panic::UnwindSafe;
 #[cfg(not(target_arch = "wasm32"))]
@@ -490,16 +493,20 @@ impl GraphData {
 
     /// Load the model input from a file
     pub fn from_path(path: std::path::PathBuf) -> Result<Self, Box<dyn std::error::Error>> {
-        let mut file = std::fs::File::open(path.clone())
-            .map_err(|_| format!("failed to open input at {}", path.display()))?;
-        let mut data = String::new();
-        file.read_to_string(&mut data)?;
-        serde_json::from_str(&data).map_err(|e| e.into())
+        let reader = std::fs::File::open(path)?;
+        let mut reader = BufReader::with_capacity(*EZKL_BUF_CAPACITY, reader);
+        let mut buf = String::new();
+        reader.read_to_string(&mut buf)?;
+        let graph_input = serde_json::from_str(&buf)?;
+        Ok(graph_input)
     }
 
     /// Save the model input to a file
     pub fn save(&self, path: std::path::PathBuf) -> Result<(), Box<dyn std::error::Error>> {
-        serde_json::to_writer(std::fs::File::create(path)?, &self).map_err(|e| e.into())
+        // buf writer
+        let writer = BufWriter::with_capacity(*EZKL_BUF_CAPACITY, std::fs::File::create(path)?);
+        serde_json::to_writer(writer, self)?;
+        Ok(())
     }
 
     ///
@@ -617,13 +624,13 @@ impl ToPyObject for DataSource {
 }
 
 #[cfg(feature = "python-bindings")]
-use crate::pfsys::field_to_string_montgomery;
+use crate::pfsys::field_to_string;
 
 #[cfg(feature = "python-bindings")]
 impl ToPyObject for FileSourceInner {
     fn to_object(&self, py: Python) -> PyObject {
         match self {
-            FileSourceInner::Field(data) => field_to_string_montgomery(data).to_object(py),
+            FileSourceInner::Field(data) => field_to_string(data).to_object(py),
             FileSourceInner::Bool(data) => data.to_object(py),
             FileSourceInner::Float(data) => data.to_object(py),
         }

src/graph/mod.rs

@@ -16,6 +16,7 @@ use halo2_proofs::plonk::VerifyingKey;
 use halo2_proofs::poly::kzg::commitment::ParamsKZG;
 pub use input::DataSource;
 use itertools::Itertools;
+use tosubcommand::ToFlags;
 
 #[cfg(not(target_arch = "wasm32"))]
 use self::input::OnChainSource;
@@ -23,19 +24,22 @@ use self::input::{FileSource, GraphData};
 use self::modules::{GraphModules, ModuleConfigs, ModuleForwardResult, ModuleSizes};
 use crate::circuit::lookup::LookupOp;
 use crate::circuit::modules::ModulePlanner;
-use crate::circuit::table::{Range, Table, RESERVED_BLINDING_ROWS_PAD};
+use crate::circuit::table::{num_cols_required, Range, Table, RESERVED_BLINDING_ROWS_PAD};
 use crate::circuit::{CheckMode, InputType};
 use crate::fieldutils::felt_to_f64;
 use crate::pfsys::PrettyElements;
 use crate::tensor::{Tensor, ValTensor};
-use crate::RunArgs;
+use crate::{RunArgs, EZKL_BUF_CAPACITY};
+
 use halo2_proofs::{
     circuit::Layouter,
     plonk::{Circuit, ConstraintSystem, Error as PlonkError},
 };
 use halo2curves::bn256::{self, Bn256, Fr as Fp, G1Affine};
 use halo2curves::ff::PrimeField;
-use log::{debug, error, info, trace, warn};
+#[cfg(not(target_arch = "wasm32"))]
+use lazy_static::lazy_static;
+use log::{debug, error, trace, warn};
 use maybe_rayon::prelude::{IntoParallelRefIterator, ParallelIterator};
 pub use model::*;
 pub use node::*;
@@ -46,20 +50,36 @@ use pyo3::types::PyDict;
 #[cfg(feature = "python-bindings")]
 use pyo3::ToPyObject;
 use serde::{Deserialize, Serialize};
-use std::io::{Read, Write};
 use std::ops::Deref;
 use thiserror::Error;
 pub use utilities::*;
 pub use vars::*;
 
 #[cfg(feature = "python-bindings")]
-use crate::pfsys::field_to_string_montgomery;
+use crate::pfsys::field_to_string;
 
 /// The safety factor for the range of the lookup table.
 pub const RANGE_MULTIPLIER: i128 = 2;
 
+/// The maximum number of columns in a lookup table.
+pub const MAX_NUM_LOOKUP_COLS: usize = 12;
+
 /// Max representation of a lookup table input
-pub const MAX_LOOKUP_ABS: i128 = 8 * 2_i128.pow(MAX_PUBLIC_SRS);
+pub const MAX_LOOKUP_ABS: i128 = (MAX_NUM_LOOKUP_COLS as i128) * 2_i128.pow(MAX_PUBLIC_SRS);
+
+#[cfg(not(target_arch = "wasm32"))]
+lazy_static! {
+    /// Max circuit area
+    pub static ref EZKL_MAX_CIRCUIT_AREA: Option<usize> =
+        if let Ok(max_circuit_area) = std::env::var("EZKL_MAX_CIRCUIT_AREA") {
+            Some(max_circuit_area.parse().unwrap_or(0))
+        } else {
+            None
+        };
+}
+
+#[cfg(target_arch = "wasm32")]
+const EZKL_MAX_CIRCUIT_AREA: Option<usize> = None;
 
 /// circuit related errors.
 #[derive(Debug, Error)]
@@ -117,15 +137,16 @@ pub enum GraphError {
     MissingResults,
 }
 
-const ASSUMED_BLINDING_FACTORS: usize = 5;
+///
+pub const ASSUMED_BLINDING_FACTORS: usize = 5;
 /// The minimum number of rows in the grid
 pub const MIN_LOGROWS: u32 = 6;
 
 /// 26
 pub const MAX_PUBLIC_SRS: u32 = bn256::Fr::S - 2;
 
-/// Lookup deg
-pub const LOOKUP_DEG: usize = 5;
+///
+pub const RESERVED_BLINDING_ROWS: usize = ASSUMED_BLINDING_FACTORS + RESERVED_BLINDING_ROWS_PAD;
 
 use std::cell::RefCell;
 
@@ -154,6 +175,8 @@ pub struct GraphWitness {
     pub max_lookup_inputs: i128,
     /// max lookup input
     pub min_lookup_inputs: i128,
+    /// max range check size
+    pub max_range_size: i128,
 }
 
 impl GraphWitness {
@@ -181,6 +204,7 @@ impl GraphWitness {
             processed_outputs: None,
             max_lookup_inputs: 0,
             min_lookup_inputs: 0,
+            max_range_size: 0,
         }
     }
 
@@ -191,42 +215,41 @@ impl GraphWitness {
         output_scales: Vec<crate::Scale>,
         visibility: VarVisibility,
     ) {
-        let mut pretty_elements = PrettyElements::default();
-        pretty_elements.rescaled_inputs = self
-            .inputs
-            .iter()
-            .enumerate()
-            .map(|(i, t)| {
-                let scale = input_scales[i];
-                t.iter()
-                    .map(|x| dequantize(*x, scale, 0.).to_string())
-                    .collect()
-            })
-            .collect();
-
-        pretty_elements.inputs = self
-            .inputs
-            .iter()
-            .map(|t| t.iter().map(|x| format!("{:?}", x)).collect())
-            .collect();
-
-        pretty_elements.rescaled_outputs = self
-            .outputs
-            .iter()
-            .enumerate()
-            .map(|(i, t)| {
-                let scale = output_scales[i];
-                t.iter()
-                    .map(|x| dequantize(*x, scale, 0.).to_string())
-                    .collect()
-            })
-            .collect();
-
-        pretty_elements.outputs = self
-            .outputs
-            .iter()
-            .map(|t| t.iter().map(|x| format!("{:?}", x)).collect())
-            .collect();
+        let mut pretty_elements = PrettyElements {
+            rescaled_inputs: self
+                .inputs
+                .iter()
+                .enumerate()
+                .map(|(i, t)| {
+                    let scale = input_scales[i];
+                    t.iter()
+                        .map(|x| dequantize(*x, scale, 0.).to_string())
+                        .collect()
+                })
+                .collect(),
+            inputs: self
+                .inputs
+                .iter()
+                .map(|t| t.iter().map(|x| format!("{:?}", x)).collect())
+                .collect(),
+            rescaled_outputs: self
+                .outputs
+                .iter()
+                .enumerate()
+                .map(|(i, t)| {
+                    let scale = output_scales[i];
+                    t.iter()
+                        .map(|x| dequantize(*x, scale, 0.).to_string())
+                        .collect()
+                })
+                .collect(),
+            outputs: self
+                .outputs
+                .iter()
+                .map(|t| t.iter().map(|x| format!("{:?}", x)).collect())
+                .collect(),
+            ..Default::default()
+        };
 
         if let Some(processed_inputs) = self.processed_inputs.clone() {
             pretty_elements.processed_inputs = processed_inputs
@@ -292,16 +315,20 @@ impl GraphWitness {
 
     /// Load the model input from a file
     pub fn from_path(path: std::path::PathBuf) -> Result<Self, Box<dyn std::error::Error>> {
-        let mut file = std::fs::File::open(path.clone())
+        let file = std::fs::File::open(path.clone())
             .map_err(|_| format!("failed to load model at {}", path.display()))?;
-        let mut data = String::new();
-        file.read_to_string(&mut data)?;
-        serde_json::from_str(&data).map_err(|e| e.into())
+
+        let reader = std::io::BufReader::with_capacity(*EZKL_BUF_CAPACITY, file);
+        serde_json::from_reader(reader).map_err(|e| e.into())
     }
 
     /// Save the model input to a file
     pub fn save(&self, path: std::path::PathBuf) -> Result<(), Box<dyn std::error::Error>> {
-        serde_json::to_writer(std::fs::File::create(path)?, &self).map_err(|e| e.into())
+        // use buf writer
+        let writer =
+            std::io::BufWriter::with_capacity(*EZKL_BUF_CAPACITY, std::fs::File::create(path)?);
+
+        serde_json::to_writer(writer, &self).map_err(|e| e.into())
     }
 
     ///
@@ -335,19 +362,23 @@ impl ToPyObject for GraphWitness {
         let inputs: Vec<Vec<String>> = self
             .inputs
             .iter()
-            .map(|x| x.iter().map(field_to_string_montgomery).collect())
+            .map(|x| x.iter().map(field_to_string).collect())
             .collect();
 
         let outputs: Vec<Vec<String>> = self
             .outputs
             .iter()
-            .map(|x| x.iter().map(field_to_string_montgomery).collect())
+            .map(|x| x.iter().map(field_to_string).collect())
             .collect();
 
         dict.set_item("inputs", inputs).unwrap();
         dict.set_item("outputs", outputs).unwrap();
         dict.set_item("max_lookup_inputs", self.max_lookup_inputs)
             .unwrap();
+        dict.set_item("min_lookup_inputs", self.min_lookup_inputs)
+            .unwrap();
+        dict.set_item("max_range_size", self.max_range_size)
+            .unwrap();
 
         if let Some(processed_inputs) = &self.processed_inputs {
             //poseidon_hash
@@ -389,10 +420,7 @@ impl ToPyObject for GraphWitness {
 
 #[cfg(feature = "python-bindings")]
 fn insert_poseidon_hash_pydict(pydict: &PyDict, poseidon_hash: &Vec<Fp>) -> Result<(), PyErr> {
-    let poseidon_hash: Vec<String> = poseidon_hash
-        .iter()
-        .map(field_to_string_montgomery)
-        .collect();
+    let poseidon_hash: Vec<String> = poseidon_hash.iter().map(field_to_string).collect();
     pydict.set_item("poseidon_hash", poseidon_hash)?;
 
     Ok(())
@@ -444,6 +472,20 @@ pub struct GraphSettings {
 }
 
 impl GraphSettings {
+    fn model_constraint_logrows(&self) -> u32 {
+        (self.num_rows as f64 + RESERVED_BLINDING_ROWS as f64)
+            .log2()
+            .ceil() as u32
+    }
+
+    fn module_constraint_logrows(&self) -> u32 {
+        (self.module_sizes.max_constraints() as f64).log2().ceil() as u32
+    }
+
+    fn constants_logrows(&self) -> u32 {
+        (self.total_const_size as f64).log2().ceil() as u32
+    }
+
     /// calculate the total number of instances
     pub fn total_instances(&self) -> Vec<usize> {
         let mut instances: Vec<usize> = self
@@ -456,22 +498,33 @@ impl GraphSettings {
         instances
     }
 
+    /// calculate the log2 of the total number of instances
+    pub fn log2_total_instances(&self) -> u32 {
+        let sum = self.total_instances().iter().sum::<usize>();
+
+        // max between 1 and the log2 of the sums
+        std::cmp::max((sum as f64).log2().ceil() as u32, 1)
+    }
+
     /// save params to file
     pub fn save(&self, path: &std::path::PathBuf) -> Result<(), std::io::Error> {
-        let encoded = serde_json::to_string(&self)?;
-        let mut file = std::fs::File::create(path)?;
-        file.write_all(encoded.as_bytes())
+        // buf writer
+        let writer =
+            std::io::BufWriter::with_capacity(*EZKL_BUF_CAPACITY, std::fs::File::create(path)?);
+        serde_json::to_writer(writer, &self).map_err(|e| {
+            error!("failed to save settings file at {}", e);
+            std::io::Error::new(std::io::ErrorKind::Other, e)
+        })
     }
     /// load params from file
     pub fn load(path: &std::path::PathBuf) -> Result<Self, std::io::Error> {
-        let mut file = std::fs::File::open(path).map_err(|e| {
-            error!("failed to open settings file at {}", e);
-            e
-        })?;
-        let mut data = String::new();
-        file.read_to_string(&mut data)?;
-        let res = serde_json::from_str(&data)?;
-        Ok(res)
+        // buf reader
+        let reader =
+            std::io::BufReader::with_capacity(*EZKL_BUF_CAPACITY, std::fs::File::open(path)?);
+        serde_json::from_reader(reader).map_err(|e| {
+            error!("failed to load settings file at {}", e);
+            std::io::Error::new(std::io::ErrorKind::Other, e)
+        })
     }
 
     /// Export the ezkl configuration as json
@@ -530,6 +583,7 @@ impl GraphSettings {
 pub struct GraphConfig {
     model_config: ModelConfig,
     module_configs: ModuleConfigs,
+    circuit_size: CircuitSize,
 }
 
 /// Defines the circuit for a computational graph / model loaded from a `.onnx` file.
@@ -566,7 +620,7 @@ impl GraphCircuit {
     ///
     pub fn save(&self, path: std::path::PathBuf) -> Result<(), Box<dyn std::error::Error>> {
         let f = std::fs::File::create(path)?;
-        let writer = std::io::BufWriter::new(f);
+        let writer = std::io::BufWriter::with_capacity(*EZKL_BUF_CAPACITY, f);
         bincode::serialize_into(writer, &self)?;
         Ok(())
     }
@@ -574,11 +628,10 @@ impl GraphCircuit {
     ///
     pub fn load(path: std::path::PathBuf) -> Result<Self, Box<dyn std::error::Error>> {
         // read bytes from file
-        let mut f = std::fs::File::open(&path)?;
-        let metadata = std::fs::metadata(&path)?;
-        let mut buffer = vec![0; metadata.len() as usize];
-        f.read_exact(&mut buffer)?;
-        let result = bincode::deserialize(&buffer)?;
+        let f = std::fs::File::open(path)?;
+        let reader = std::io::BufReader::with_capacity(*EZKL_BUF_CAPACITY, f);
+        let result: GraphCircuit = bincode::deserialize_from(reader)?;
+
         Ok(result)
     }
 }
@@ -593,6 +646,17 @@ pub enum TestDataSource {
     OnChain,
 }
 
+impl std::fmt::Display for TestDataSource {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        match self {
+            TestDataSource::File => write!(f, "file"),
+            TestDataSource::OnChain => write!(f, "on-chain"),
+        }
+    }
+}
+
+impl ToFlags for TestDataSource {}
+
 impl From<String> for TestDataSource {
     fn from(value: String) -> Self {
         match value.to_lowercase().as_str() {
@@ -809,7 +873,7 @@ impl GraphCircuit {
         let shapes = self.model().graph.input_shapes()?;
         let scales = self.model().graph.get_input_scales();
         let input_types = self.model().graph.get_input_types()?;
-        info!("input scales: {:?}", scales);
+        debug!("input scales: {:?}", scales);
 
         match &data.input_data {
             DataSource::File(file_data) => {
@@ -828,7 +892,7 @@ impl GraphCircuit {
         let shapes = self.model().graph.input_shapes()?;
         let scales = self.model().graph.get_input_scales();
         let input_types = self.model().graph.get_input_types()?;
-        info!("input scales: {:?}", scales);
+        debug!("input scales: {:?}", scales);
 
         self.process_data_source(&data.input_data, shapes, scales, input_types)
             .await
@@ -954,18 +1018,10 @@ impl GraphCircuit {
         Ok(data)
     }
 
-    fn reserved_blinding_rows() -> f64 {
-        (ASSUMED_BLINDING_FACTORS + RESERVED_BLINDING_ROWS_PAD) as f64
-    }
-
-    fn calc_safe_lookup_range(
-        min_lookup_inputs: i128,
-        max_lookup_inputs: i128,
-        lookup_safety_margin: i128,
-    ) -> Range {
+    fn calc_safe_lookup_range(min_max_lookup: Range, lookup_safety_margin: i128) -> Range {
         let mut margin = (
-            lookup_safety_margin * min_lookup_inputs,
-            lookup_safety_margin * max_lookup_inputs,
+            lookup_safety_margin * min_max_lookup.0,
+            lookup_safety_margin * min_max_lookup.1,
         );
         if lookup_safety_margin == 1 {
             margin.0 += 4;
@@ -975,18 +1031,33 @@ impl GraphCircuit {
         margin
     }
 
-    fn calc_num_cols(safe_range: Range, max_logrows: u32) -> usize {
-        let max_col_size = Table::<Fp>::cal_col_size(
-            max_logrows as usize,
-            Self::reserved_blinding_rows() as usize,
+    fn calc_num_cols(range_len: i128, max_logrows: u32) -> usize {
+        let max_col_size = Table::<Fp>::cal_col_size(max_logrows as usize, RESERVED_BLINDING_ROWS);
+        num_cols_required(range_len, max_col_size)
+    }
+
+    fn table_size_logrows(
+        &self,
+        safe_lookup_range: Range,
+        max_range_size: i128,
+    ) -> Result<u32, Box<dyn std::error::Error>> {
+        // pick the range with the largest absolute size safe_lookup_range or max_range_size
+        let safe_range = std::cmp::max(
+            (safe_lookup_range.1 - safe_lookup_range.0).abs(),
+            max_range_size,
         );
-        Table::<Fp>::num_cols_required(safe_range, max_col_size)
+
+        let min_bits = (safe_range as f64 + RESERVED_BLINDING_ROWS as f64 + 1.)
+            .log2()
+            .ceil() as u32;
+
+        Ok(min_bits)
     }
 
     fn calc_min_logrows(
         &mut self,
-        min_lookup_inputs: i128,
-        max_lookup_inputs: i128,
+        min_max_lookup: Range,
+        max_range_size: i128,
         max_logrows: Option<u32>,
         lookup_safety_margin: i128,
     ) -> Result<(), Box<dyn std::error::Error>> {
@@ -996,125 +1067,88 @@ impl GraphCircuit {
         let mut max_logrows = std::cmp::max(max_logrows, MIN_LOGROWS);
         let mut min_logrows = MIN_LOGROWS;
 
-        let reserved_blinding_rows = Self::reserved_blinding_rows();
+        let safe_lookup_range = Self::calc_safe_lookup_range(min_max_lookup, lookup_safety_margin);
+
         // check if has overflowed max lookup input
-        if max_lookup_inputs > MAX_LOOKUP_ABS / lookup_safety_margin
-            || min_lookup_inputs < -MAX_LOOKUP_ABS / lookup_safety_margin
-        {
-            let err_string = format!("max lookup input ({}) is too large", max_lookup_inputs);
+        if (min_max_lookup.1 - min_max_lookup.0).abs() > MAX_LOOKUP_ABS / lookup_safety_margin {
+            let err_string = format!("max lookup input {:?} is too large", min_max_lookup);
             return Err(err_string.into());
         }
 
-        let safe_range = Self::calc_safe_lookup_range(
-            min_lookup_inputs,
-            max_lookup_inputs,
-            lookup_safety_margin,
+        if max_range_size.abs() > MAX_LOOKUP_ABS {
+            let err_string = format!("max range check size {:?} is too large", max_range_size);
+            return Err(err_string.into());
+        }
+
+        // These are hard lower limits, we can't overflow instances or modules constraints
+        let instance_logrows = self.settings().log2_total_instances();
+        let module_constraint_logrows = self.settings().module_constraint_logrows();
+        min_logrows = std::cmp::max(
+            min_logrows,
+            // max of the instance logrows and the module constraint logrows is the lower limit
+            [instance_logrows, module_constraint_logrows]
+                .iter()
+                .max()
+                .unwrap()
+                .clone(),
         );
 
+        // These are upper limits, going above these is wasteful, but they are not hard limits
+        let model_constraint_logrows = self.settings().model_constraint_logrows();
+        let min_bits = self.table_size_logrows(safe_lookup_range, max_range_size)?;
+        let constants_logrows = self.settings().constants_logrows();
+        max_logrows = std::cmp::min(
+            max_logrows,
+            // max of the model constraint logrows, min_bits, and the constants logrows is the upper limit
+            [model_constraint_logrows, min_bits, constants_logrows]
+                .iter()
+                .max()
+                .unwrap()
+                .clone(),
+        );
+
+        // we now have a min and max logrows
+        max_logrows = std::cmp::max(min_logrows, max_logrows);
+
         // degrade the max logrows until the extended k is small enough
         while min_logrows < max_logrows
-            && !self.extended_k_is_small_enough(
-                min_logrows,
-                Self::calc_num_cols(safe_range, min_logrows),
-            )
-        {
-            min_logrows += 1;
-        }
-
-        if !self
-            .extended_k_is_small_enough(min_logrows, Self::calc_num_cols(safe_range, min_logrows))
-        {
-            let err_string = format!(
-                "extended k is too large to accommodate the quotient polynomial with logrows {}",
-                min_logrows
-            );
-            error!("{}", err_string);
-            return Err(err_string.into());
-        }
-
-        while min_logrows < max_logrows
-            && !self.extended_k_is_small_enough(
-                max_logrows,
-                Self::calc_num_cols(safe_range, max_logrows),
-            )
+            && !self.extended_k_is_small_enough(max_logrows, safe_lookup_range, max_range_size)
         {
             max_logrows -= 1;
         }
 
-        if !self
-            .extended_k_is_small_enough(max_logrows, Self::calc_num_cols(safe_range, max_logrows))
-        {
+        if !self.extended_k_is_small_enough(max_logrows, safe_lookup_range, max_range_size) {
             let err_string = format!(
                 "extended k is too large to accommodate the quotient polynomial with logrows {}",
                 max_logrows
             );
-            error!("{}", err_string);
+            debug!("{}", err_string);
             return Err(err_string.into());
         }
 
-        let min_bits = ((safe_range.1 - safe_range.0) as f64 + reserved_blinding_rows + 1.)
-            .log2()
-            .ceil() as usize;
-
-        let min_rows_from_constraints = (self.settings().num_rows as f64 + reserved_blinding_rows)
-            .log2()
-            .ceil() as usize;
-
-        let mut logrows = std::cmp::max(min_bits, min_rows_from_constraints);
-
-        // if public input then public inputs col will have public inputs len
-        if self.settings().run_args.input_visibility.is_public()
-            || self.settings().run_args.output_visibility.is_public()
-        {
-            let mut max_instance_len = self
-                .model()
-                .instance_shapes()?
-                .iter()
-                .fold(0, |acc, x| std::cmp::max(acc, x.iter().product::<usize>()))
-                as f64
-                + reserved_blinding_rows;
-            // if there are modules then we need to add the max module size
-            if self.settings().uses_modules() {
-                max_instance_len += self
-                    .settings()
-                    .module_sizes
-                    .num_instances()
-                    .iter()
-                    .sum::<usize>() as f64;
-            }
-            let instance_len_logrows = (max_instance_len).log2().ceil() as usize;
-            logrows = std::cmp::max(logrows, instance_len_logrows);
-            // this is for fixed const columns
-        }
-
-        // ensure logrows is at least 4
-        logrows = std::cmp::max(logrows, min_logrows as usize);
-        logrows = std::cmp::min(logrows, max_logrows as usize);
+        let logrows = max_logrows;
 
         let model = self.model().clone();
         let settings_mut = self.settings_mut();
-        settings_mut.run_args.lookup_range = safe_range;
-        settings_mut.run_args.logrows = logrows as u32;
+        settings_mut.run_args.lookup_range = safe_lookup_range;
+        settings_mut.run_args.logrows = logrows;
 
         *settings_mut = GraphCircuit::new(model, &settings_mut.run_args)?
             .settings()
             .clone();
 
-        // recalculate the total const size give nthe new logrows
-        let total_const_len = settings_mut.total_const_size;
-        let const_len_logrows = (total_const_len as f64).log2().ceil() as u32;
-        settings_mut.run_args.logrows =
-            std::cmp::max(settings_mut.run_args.logrows, const_len_logrows);
-        // recalculate the total number of constraints given the new logrows
-        let min_rows_from_constraints = (settings_mut.num_rows as f64 + reserved_blinding_rows)
-            .log2()
-            .ceil() as u32;
-        settings_mut.run_args.logrows =
-            std::cmp::max(settings_mut.run_args.logrows, min_rows_from_constraints);
+        // recalculate the logrows if there has been overflow on the constants
+        settings_mut.run_args.logrows = std::cmp::max(
+            settings_mut.run_args.logrows,
+            settings_mut.constants_logrows(),
+        );
+        // recalculate the logrows if there has been overflow for the model constraints
+        settings_mut.run_args.logrows = std::cmp::max(
+            settings_mut.run_args.logrows,
+            settings_mut.model_constraint_logrows(),
+        );
 
-        settings_mut.run_args.logrows = std::cmp::min(max_logrows, settings_mut.run_args.logrows);
-
-        info!(
+        debug!(
             "setting lookup_range to: {:?}, setting logrows to: {}",
             self.settings().run_args.lookup_range,
             self.settings().run_args.logrows
@@ -1123,12 +1157,37 @@ impl GraphCircuit {
         Ok(())
     }
 
-    fn extended_k_is_small_enough(&self, k: u32, num_lookup_cols: usize) -> bool {
-        let max_degree = self.settings().run_args.num_inner_cols + 2;
-        let max_lookup_degree = LOOKUP_DEG + num_lookup_cols - 1; // num_lookup_cols - 1 is the degree of the lookup synthetic selector
+    fn extended_k_is_small_enough(
+        &self,
+        k: u32,
+        safe_lookup_range: Range,
+        max_range_size: i128,
+    ) -> bool {
+        // if num cols is too large then the extended k is too large
+        if Self::calc_num_cols(safe_lookup_range.1 - safe_lookup_range.0, k) > MAX_NUM_LOOKUP_COLS {
+            return false;
+        } else if Self::calc_num_cols(max_range_size, k) > MAX_NUM_LOOKUP_COLS {
+            return false;
+        }
 
-        let max_degree = std::cmp::max(max_degree, max_lookup_degree);
+        let mut settings = self.settings().clone();
+        settings.run_args.lookup_range = safe_lookup_range;
+        settings.run_args.logrows = k;
+        settings.required_range_checks = vec![(0, max_range_size)];
+        let mut cs = ConstraintSystem::default();
+        // fetch gag
+        #[cfg(unix)]
+        let _r = match gag::Gag::stdout() {
+            Ok(r) => Some(r),
+            Err(_) => None,
+        };
+        Self::configure_with_params(&mut cs, settings);
+        #[cfg(feature = "mv-lookup")]
+        let cs = cs.chunk_lookups();
         // quotient_poly_degree * params.n - 1 is the degree of the quotient polynomial
+        let max_degree = cs.degree();
+        #[cfg(unix)]
+        std::mem::drop(_r);
         let quotient_poly_degree = (max_degree - 1) as u64;
         // n = 2^k
         let n = 1u64 << k;
@@ -1146,14 +1205,14 @@ impl GraphCircuit {
     /// Calibrate the circuit to the supplied data.
     pub fn calibrate_from_min_max(
         &mut self,
-        min_lookup_inputs: i128,
-        max_lookup_inputs: i128,
+        min_max_lookup: Range,
+        max_range_size: i128,
         max_logrows: Option<u32>,
         lookup_safety_margin: i128,
     ) -> Result<(), Box<dyn std::error::Error>> {
         self.calc_min_logrows(
-            min_lookup_inputs,
-            max_lookup_inputs,
+            min_max_lookup,
+            max_range_size,
             max_logrows,
             lookup_safety_margin,
         )?;
@@ -1166,6 +1225,7 @@ impl GraphCircuit {
         inputs: &mut [Tensor<Fp>],
         vk: Option<&VerifyingKey<G1Affine>>,
         srs: Option<&ParamsKZG<Bn256>>,
+        throw_range_check_error: bool,
     ) -> Result<GraphWitness, Box<dyn std::error::Error>> {
         let original_inputs = inputs.to_vec();
 
@@ -1206,7 +1266,9 @@ impl GraphCircuit {
             }
         }
 
-        let mut model_results = self.model().forward(inputs)?;
+        let mut model_results =
+            self.model()
+                .forward(inputs, &self.settings().run_args, throw_range_check_error)?;
 
         if visibility.output.requires_processing() {
             let module_outlets = visibility.output.overwrites_inputs();
@@ -1249,6 +1311,7 @@ impl GraphCircuit {
             processed_outputs,
             max_lookup_inputs: model_results.max_lookup_inputs,
             min_lookup_inputs: model_results.min_lookup_inputs,
+            max_range_size: model_results.max_range_size,
         };
 
         witness.generate_rescaled_elements(
@@ -1366,7 +1429,6 @@ impl GraphCircuit {
     }
 }
 
-#[cfg(not(target_arch = "wasm32"))]
 #[derive(Clone, Debug, Default, Serialize, Deserialize)]
 struct CircuitSize {
     num_instances: usize,
@@ -1374,20 +1436,22 @@ struct CircuitSize {
     num_fixed: usize,
     num_challenges: usize,
     num_selectors: usize,
+    logrows: u32,
 }
 
-#[cfg(not(target_arch = "wasm32"))]
 impl CircuitSize {
-    pub fn from_cs(cs: &ConstraintSystem<Fp>) -> Self {
+    pub fn from_cs(cs: &ConstraintSystem<Fp>, logrows: u32) -> Self {
         CircuitSize {
             num_instances: cs.num_instance_columns(),
             num_advice_columns: cs.num_advice_columns(),
             num_fixed: cs.num_fixed_columns(),
             num_challenges: cs.num_challenges(),
             num_selectors: cs.num_selectors(),
+            logrows,
         }
     }
 
+    #[cfg(not(target_arch = "wasm32"))]
     /// Export the ezkl configuration as json
     pub fn as_json(&self) -> Result<String, Box<dyn std::error::Error>> {
         let serialized = match serde_json::to_string(&self) {
@@ -1398,6 +1462,25 @@ impl CircuitSize {
         };
         Ok(serialized)
     }
+
+    /// number of columns
+    pub fn num_columns(&self) -> usize {
+        self.num_instances + self.num_advice_columns + self.num_fixed
+    }
+
+    /// area of the circuit
+    pub fn area(&self) -> usize {
+        self.num_columns() * (1 << self.logrows)
+    }
+
+    /// area less than max
+    pub fn area_less_than_max(&self) -> bool {
+        if EZKL_MAX_CIRCUIT_AREA.is_some() {
+            self.area() < EZKL_MAX_CIRCUIT_AREA.unwrap()
+        } else {
+            true
+        }
+    }
 }
 
 impl Circuit<Fp> for GraphCircuit {
@@ -1472,10 +1555,12 @@ impl Circuit<Fp> for GraphCircuit {
             (cs.degree() as f32).log2().ceil()
         );
 
+        let circuit_size = CircuitSize::from_cs(cs, params.run_args.logrows);
+
         #[cfg(not(target_arch = "wasm32"))]
-        info!(
+        debug!(
             "circuit size: \n {}",
-            CircuitSize::from_cs(cs)
+            circuit_size
                 .as_json()
                 .unwrap()
                 .to_colored_json_auto()
@@ -1485,6 +1570,7 @@ impl Circuit<Fp> for GraphCircuit {
         GraphConfig {
             model_config,
             module_configs,
+            circuit_size,
         }
     }
 
@@ -1497,6 +1583,16 @@ impl Circuit<Fp> for GraphCircuit {
         config: Self::Config,
         mut layouter: impl Layouter<Fp>,
     ) -> Result<(), PlonkError> {
+        // check if the circuit area is less than the max
+        if !config.circuit_size.area_less_than_max() {
+            error!(
+                "circuit area {} is larger than the max allowed area {}",
+                config.circuit_size.area(),
+                EZKL_MAX_CIRCUIT_AREA.unwrap()
+            );
+            return Err(PlonkError::Synthesis);
+        }
+
         trace!("Setting input in synthesize");
         let input_vis = &self.settings().run_args.input_visibility;
         let output_vis = &self.settings().run_args.output_visibility;

src/graph/model.rs

@@ -10,7 +10,6 @@ use crate::circuit::table::Range;
 use crate::circuit::Input;
 use crate::circuit::InputType;
 use crate::circuit::Unknown;
-use crate::fieldutils::felt_to_i128;
 use crate::tensor::ValType;
 use crate::{
     circuit::{lookup::LookupOp, BaseConfig as PolyConfig, CheckMode, Op},
@@ -57,6 +56,8 @@ use unzip_n::unzip_n;
 
 unzip_n!(pub 3);
 
+#[cfg(not(target_arch = "wasm32"))]
+type TractResult = (Graph<TypedFact, Box<dyn TypedOp>>, SymbolValues);
 /// The result of a forward pass.
 #[derive(Clone, Debug)]
 pub struct ForwardResult {
@@ -66,6 +67,19 @@ pub struct ForwardResult {
     pub max_lookup_inputs: i128,
     /// The minimum value of any input to a lookup operation.
     pub min_lookup_inputs: i128,
+    /// The max range check size
+    pub max_range_size: 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,
+            max_range_size: res.max_range_size,
+        }
+    }
 }
 
 /// A circuit configuration for the entirety of a model loaded from an Onnx file.
@@ -93,6 +107,14 @@ 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,
+    /// min range check
+    pub max_range_size: i128,
+    /// outputs
+    pub outputs: Vec<Tensor<Fp>>,
 }
 
 /// A struct for loading from an Onnx file and converting a computational graph to a circuit.
@@ -478,7 +500,7 @@ impl Model {
     ) -> Result<GraphSettings, Box<dyn Error>> {
         let instance_shapes = self.instance_shapes()?;
         #[cfg(not(target_arch = "wasm32"))]
-        info!(
+        debug!(
             "{} {} {}",
             "model has".blue(),
             instance_shapes.len().to_string().blue(),
@@ -486,7 +508,25 @@ impl Model {
         );
         // this is the total number of variables we will need to allocate
         // for the circuit
-        let res = self.dummy_layout(run_args, &self.graph.input_shapes()?)?;
+        let default_value = if !self.visibility.input.is_fixed() {
+            ValType::Value(Value::<Fp>::unknown())
+        } else {
+            ValType::Constant(Fp::ONE)
+        };
+
+        let inputs: Vec<ValTensor<Fp>> = self
+            .graph
+            .input_shapes()?
+            .iter()
+            .map(|shape| {
+                let mut t: ValTensor<Fp> =
+                    vec![default_value.clone(); shape.iter().product()].into();
+                t.reshape(shape)?;
+                Ok(t)
+            })
+            .collect::<Result<Vec<_>, Box<dyn Error>>>()?;
+
+        let res = self.dummy_layout(run_args, &inputs, false)?;
 
         // if we're using percentage tolerance, we need to add the necessary range check ops for it.
 
@@ -521,207 +561,18 @@ impl Model {
     /// * `reader` - A reader for an Onnx file.
     /// * `model_inputs` - A vector of [Tensor]s to use as inputs to the model.
     /// * `run_args` - [RunArgs]
-    pub fn forward(&self, model_inputs: &[Tensor<Fp>]) -> Result<ForwardResult, Box<dyn Error>> {
-        let mut results: BTreeMap<&usize, Vec<Tensor<Fp>>> = BTreeMap::new();
-        let mut max_lookup_inputs = 0;
-        let mut min_lookup_inputs = 0;
-
-        let input_shapes = self.graph.input_shapes()?;
-
-        for (i, input_idx) in self.graph.inputs.iter().enumerate() {
-            let mut input = model_inputs[i].clone();
-            input.reshape(&input_shapes[i])?;
-            results.insert(input_idx, vec![input]);
-        }
-
-        for (idx, n) in self.graph.nodes.iter() {
-            let mut inputs = vec![];
-            if n.is_input() {
-                let t = results.get(idx).ok_or(GraphError::MissingResults)?[0].clone();
-                inputs.push(t);
-            } else {
-                for (idx, outlet) in n.inputs().iter() {
-                    match results.get(&idx) {
-                        Some(value) => inputs.push(value[*outlet].clone()),
-                        None => return Err(Box::new(GraphError::MissingNode(*idx))),
-                    }
-                }
-            };
-
-            debug!("executing {}: {}", idx, n.as_str());
-            debug!("dims: {:?}", n.out_dims());
-            debug!(
-                "input_dims: {:?}",
-                inputs.iter().map(|x| x.dims()).collect::<Vec<_>>()
-            );
-
-            debug!("input nodes: {:?}", n.inputs());
-
-            if n.is_lookup() {
-                let (mut min, mut max) = (0, 0);
-                for i in &inputs {
-                    max = max.max(
-                        i.iter()
-                            .map(|x| felt_to_i128(*x))
-                            .max()
-                            .ok_or("missing max")?,
-                    );
-                    min = min.min(
-                        i.iter()
-                            .map(|x| felt_to_i128(*x))
-                            .min()
-                            .ok_or("missing min")?,
-                    );
-                }
-                max_lookup_inputs = max_lookup_inputs.max(max);
-                min_lookup_inputs = min_lookup_inputs.min(min);
-                debug!("max lookup inputs: {}", max);
-                debug!("min lookup inputs: {}", min);
-            }
-
-            match n {
-                NodeType::Node(n) => {
-                    // execute the op
-                    let start = instant::Instant::now();
-                    let mut res = Op::<Fp>::f(&n.opkind, &inputs)?;
-                    res.output.reshape(&n.out_dims)?;
-                    let elapsed = start.elapsed();
-                    trace!("op took: {:?}", elapsed);
-                    // see if any of the intermediate lookup calcs are the max
-                    if !res.intermediate_lookups.is_empty() {
-                        let (mut min, mut max) = (0, 0);
-                        for i in &res.intermediate_lookups {
-                            max = max.max(i.clone().into_iter().max().ok_or("missing max")?);
-                            min = min.min(i.clone().into_iter().min().ok_or("missing min")?);
-                        }
-                        max_lookup_inputs = max_lookup_inputs.max(max);
-                        min_lookup_inputs = min_lookup_inputs.min(min);
-                        debug!("intermediate max lookup inputs: {}", max);
-                        debug!("intermediate min lookup inputs: {}", min);
-                    }
-                    debug!(
-                        "------------ output node int {}: {} \n ------------ float: {} \n ------------ max: {} \n ------------ min: {} \n ------------ scale: {}",
-                        idx,
-                        res.output.map(crate::fieldutils::felt_to_i32).show(),
-                        res.output
-                            .map(|x| crate::fieldutils::felt_to_f64(x)
-                                / scale_to_multiplier(n.out_scale))
-                            .show(),
-                        res.output.clone().into_iter().map(crate::fieldutils::felt_to_i128).max().unwrap_or(0),
-                        res.output.clone().into_iter().map(crate::fieldutils::felt_to_i128).min().unwrap_or(0),
-                        n.out_scale
-                    );
-                    results.insert(idx, vec![res.output]);
-                }
-                NodeType::SubGraph {
-                    model,
-                    output_mappings,
-                    input_mappings,
-                    inputs: input_tuple,
-                    ..
-                } => {
-                    let orig_inputs = inputs.clone();
-                    let input_mappings = input_mappings.clone();
-
-                    let input_dims = inputs.iter().map(|inp| inp.dims());
-                    let num_iter = number_of_iterations(&input_mappings, input_dims.collect());
-
-                    debug!(
-                        "{} iteration(s) in a subgraph with inputs {:?} and sources {:?}",
-                        num_iter, input_tuple, model.graph.inputs
-                    );
-
-                    debug!("input_mappings: {:?}", input_mappings);
-
-                    let mut full_results: Vec<Tensor<Fp>> = vec![];
-
-                    for i in 0..num_iter {
-                        // replace the Stacked input with the current chunk iter
-                        for ((mapping, inp), og_input) in
-                            input_mappings.iter().zip(&mut inputs).zip(&orig_inputs)
-                        {
-                            if let InputMapping::Stacked { axis, chunk } = mapping {
-                                let start = i * chunk;
-                                let end = (i + 1) * chunk;
-                                let t = crate::tensor::ops::slice(og_input, axis, &start, &end)?;
-                                *inp = t;
-                            }
-                        }
-
-                        let res = model.forward(&inputs)?;
-                        // recursively get the max lookup inputs for subgraphs
-                        max_lookup_inputs = max_lookup_inputs.max(res.max_lookup_inputs);
-                        min_lookup_inputs = min_lookup_inputs.min(res.min_lookup_inputs);
-
-                        let mut outlets = BTreeMap::new();
-                        for (mappings, outlet_res) in output_mappings.iter().zip(res.outputs) {
-                            for mapping in mappings {
-                                match mapping {
-                                    OutputMapping::Single { outlet, .. } => {
-                                        outlets.insert(outlet, outlet_res.clone());
-                                    }
-                                    OutputMapping::Stacked { outlet, axis, .. } => {
-                                        if !full_results.is_empty() {
-                                            let stacked_res = crate::tensor::ops::concat(
-                                                &[&full_results[*outlet], &outlet_res],
-                                                *axis,
-                                            )?;
-
-                                            outlets.insert(outlet, stacked_res);
-                                        } else {
-                                            outlets.insert(outlet, outlet_res.clone());
-                                        }
-                                    }
-                                }
-                            }
-                        }
-
-                        full_results = outlets.into_values().collect_vec();
-
-                        let output_states = output_state_idx(output_mappings);
-                        let input_states = input_state_idx(&input_mappings);
-
-                        assert_eq!(input_states.len(), output_states.len());
-
-                        for (input_idx, output_idx) in input_states.iter().zip(output_states) {
-                            inputs[*input_idx] = full_results[output_idx].clone();
-                        }
-                    }
-
-                    trace!(
-                        "------------ output subgraph node {}: {:?}",
-                        idx,
-                        full_results
-                            .iter()
-                            .map(|x|
-                            // convert to tensor i32
-                            x.map(crate::fieldutils::felt_to_i32).show())
-                            .collect_vec()
-                    );
-
-                    results.insert(idx, full_results);
-                }
-            }
-        }
-
-        let output_nodes = self.graph.outputs.iter();
-        debug!(
-            "model outputs are nodes: {:?}",
-            output_nodes.clone().collect_vec()
-        );
-        let outputs = output_nodes
-            .map(|(idx, outlet)| {
-                Ok(results.get(&idx).ok_or(GraphError::MissingResults)?[*outlet].clone())
-            })
-            .collect::<Result<Vec<_>, GraphError>>()?;
-
-        let res = ForwardResult {
-            outputs,
-            max_lookup_inputs,
-            min_lookup_inputs,
-        };
-
-        Ok(res)
+    pub fn forward(
+        &self,
+        model_inputs: &[Tensor<Fp>],
+        run_args: &RunArgs,
+        throw_range_check_error: bool,
+    ) -> 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(run_args, &valtensor_inputs, throw_range_check_error)?;
+        Ok(res.into())
     }
 
     /// Loads an Onnx model from a specified path.
@@ -733,7 +584,7 @@ impl Model {
     fn load_onnx_using_tract(
         reader: &mut dyn std::io::Read,
         run_args: &RunArgs,
-    ) -> Result<(Graph<TypedFact, Box<dyn TypedOp>>, SymbolValues), Box<dyn Error>> {
+    ) -> Result<TractResult, Box<dyn Error>> {
         use tract_onnx::{
             tract_core::internal::IntoArcTensor, tract_hir::internal::GenericFactoid,
         };
@@ -772,7 +623,7 @@ impl Model {
         for (symbol, value) in run_args.variables.iter() {
             let symbol = model.symbol_table.sym(symbol);
             symbol_values = symbol_values.with(&symbol, *value as i64);
-            info!("set {} to {}", symbol, value);
+            debug!("set {} to {}", symbol, value);
         }
 
         // Note: do not optimize the model, as the layout will depend on underlying hardware
@@ -1173,14 +1024,14 @@ 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[1];
-        let index = &vars.advices[2];
+        let output = &vars.advices[2];
+        let index = &vars.advices[1];
         for op in required_lookups {
             base_gate.configure_lookup(meta, input, output, index, lookup_range, logrows, &op)?;
         }
 
         for range in required_range_checks {
-            base_gate.configure_range_check(meta, input, range)?;
+            base_gate.configure_range_check(meta, input, index, range, logrows)?;
         }
 
         Ok(base_gate)
@@ -1293,7 +1144,7 @@ impl Model {
 
         // Then number of columns in the circuits
         #[cfg(not(target_arch = "wasm32"))]
-        info!(
+        debug!(
             "{} {} {} (coord={}, constants={})",
             "model uses".blue(),
             num_rows.to_string().blue(),
@@ -1335,18 +1186,29 @@ impl Model {
             };
 
             debug!(
-                "laying out {}: {}, row:{}, coord:{}, total_constants: {}",
+                "laying out {}: {}, row:{}, coord:{}, total_constants: {}, max_lookup_inputs: {}, min_lookup_inputs: {}",
                 idx,
                 node.as_str(),
                 region.row(),
                 region.linear_coord(),
-                region.total_constants()
+                region.total_constants(),
+                region.max_lookup_inputs(),
+                region.min_lookup_inputs()
             );
             debug!("dims: {:?}", node.out_dims());
             debug!(
                 "input_dims {:?}",
                 values.iter().map(|v| v.dims()).collect_vec()
             );
+            debug!("output scales: {:?}", node.out_scales());
+            debug!("input indices: {:?}", node.inputs());
+            debug!(
+                "input scales: {:?}",
+                node.inputs()
+                    .iter()
+                    .map(|(idx, outlet)| self.graph.nodes[idx].out_scales()[*outlet])
+                    .collect_vec()
+            );
 
             match &node {
                 NodeType::Node(n) => {
@@ -1489,28 +1351,14 @@ impl Model {
     pub fn dummy_layout(
         &self,
         run_args: &RunArgs,
-        input_shapes: &[Vec<usize>],
+        inputs: &[ValTensor<Fp>],
+        throw_range_check_error: bool,
     ) -> Result<DummyPassRes, Box<dyn Error>> {
-        info!("calculating num of constraints using dummy model layout...");
+        debug!("calculating num of constraints using dummy model layout...");
 
         let start_time = instant::Instant::now();
 
         let mut results = BTreeMap::<usize, Vec<ValTensor<Fp>>>::new();
-        let default_value = if !self.visibility.input.is_fixed() {
-            ValType::Value(Value::<Fp>::unknown())
-        } else {
-            ValType::Constant(Fp::ONE)
-        };
-
-        let inputs: Vec<ValTensor<Fp>> = input_shapes
-            .iter()
-            .map(|shape| {
-                let mut t: ValTensor<Fp> =
-                    vec![default_value.clone(); shape.iter().product()].into();
-                t.reshape(shape)?;
-                Ok(t)
-            })
-            .collect::<Result<Vec<_>, Box<dyn Error>>>()?;
 
         for (i, input_idx) in self.graph.inputs.iter().enumerate() {
             results.insert(*input_idx, vec![inputs[i].clone()]);
@@ -1523,7 +1371,7 @@ impl Model {
             vars: ModelVars::new_dummy(),
         };
 
-        let mut region = RegionCtx::new_dummy(0, run_args.num_inner_cols);
+        let mut region = RegionCtx::new_dummy(0, run_args.num_inner_cols, throw_range_check_error);
 
         let outputs = self.layout_nodes(&mut model_config, &mut region, &mut results)?;
 
@@ -1534,27 +1382,26 @@ impl Model {
                 ValType::Constant(Fp::ONE)
             };
 
-            let comparator = outputs
+            let output_scales = self.graph.get_output_scales()?;
+            let res = outputs
                 .iter()
-                .map(|x| {
-                    let mut v: ValTensor<Fp> =
-                        vec![default_value.clone(); x.dims().iter().product::<usize>()].into();
-                    v.reshape(x.dims())?;
-                    Ok(v)
-                })
-                .collect::<Result<Vec<_>, Box<dyn Error>>>()?;
+                .enumerate()
+                .map(|(i, output)| {
+                    let mut tolerance = run_args.tolerance;
+                    tolerance.scale = scale_to_multiplier(output_scales[i]).into();
+
+                    let mut comparator: ValTensor<Fp> =
+                        vec![default_value.clone(); output.dims().iter().product::<usize>()].into();
+                    comparator.reshape(output.dims())?;
 
-            let _ = outputs
-                .into_iter()
-                .zip(comparator)
-                .map(|(o, c)| {
                     dummy_config.layout(
                         &mut region,
-                        &[o, c],
-                        Box::new(HybridOp::RangeCheck(run_args.tolerance)),
+                        &[output.clone(), comparator],
+                        Box::new(HybridOp::RangeCheck(tolerance)),
                     )
                 })
-                .collect::<Result<Vec<_>, _>>()?;
+                .collect::<Result<Vec<_>, _>>();
+            res?;
         } else if !self.visibility.output.is_private() {
             for output in &outputs {
                 region.increment_total_constants(output.num_constants());
@@ -1566,7 +1413,7 @@ impl Model {
 
         // Then number of columns in the circuits
         #[cfg(not(target_arch = "wasm32"))]
-        info!(
+        debug!(
             "{} {} {} (coord={}, constants={})",
             "model uses".blue(),
             region.row().to_string().blue(),
@@ -1575,12 +1422,24 @@ 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(),
+            max_range_size: region.max_range_size(),
+            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,
@@ -219,8 +219,6 @@ impl Op<Fp> for RebaseScale {
         let mut res = Op::<Fp>::f(&*self.inner, x)?;
         let rebase_res = Op::<Fp>::f(&self.rebase_op, &[res.output])?;
         res.output = rebase_res.output;
-        res.intermediate_lookups
-            .extend(rebase_res.intermediate_lookups);
 
         Ok(res)
     }
@@ -499,6 +497,7 @@ impl Node {
     /// * `public_params` - flag if parameters of model are public
     /// * `idx` - The node's unique identifier.
     #[cfg(not(target_arch = "wasm32"))]
+    #[allow(clippy::too_many_arguments)]
     pub fn new(
         node: OnnxNode<TypedFact, Box<dyn TypedOp>>,
         other_nodes: &mut BTreeMap<usize, super::NodeType>,

src/graph/utilities.rs

@@ -439,17 +439,16 @@ pub fn new_op_from_onnx(
             let op = load_op::<ScatterElements>(node.op(), idx, node.op().name().to_string())?;
             let axis = op.axis;
 
-            let mut op =
-                SupportedOp::Hybrid(crate::circuit::ops::hybrid::HybridOp::ScatterElements {
-                    dim: axis,
-                    constant_idx: None,
-                });
+            let mut op = SupportedOp::Linear(crate::circuit::ops::poly::PolyOp::ScatterElements {
+                dim: axis,
+                constant_idx: None,
+            });
 
             // if param_visibility.is_public() {
             if let Some(c) = inputs[1].opkind().get_mutable_constant() {
                 inputs[1].decrement_use();
                 deleted_indices.push(1);
-                op = SupportedOp::Hybrid(crate::circuit::ops::hybrid::HybridOp::ScatterElements {
+                op = SupportedOp::Linear(crate::circuit::ops::poly::PolyOp::ScatterElements {
                     dim: axis,
                     constant_idx: Some(c.raw_values.map(|x| x as usize)),
                 })
@@ -478,17 +477,16 @@ pub fn new_op_from_onnx(
             let op = load_op::<GatherElements>(node.op(), idx, node.op().name().to_string())?;
             let axis = op.axis;
 
-            let mut op =
-                SupportedOp::Hybrid(crate::circuit::ops::hybrid::HybridOp::GatherElements {
-                    dim: axis,
-                    constant_idx: None,
-                });
+            let mut op = SupportedOp::Linear(crate::circuit::ops::poly::PolyOp::GatherElements {
+                dim: axis,
+                constant_idx: None,
+            });
 
             // if param_visibility.is_public() {
             if let Some(c) = inputs[1].opkind().get_mutable_constant() {
                 inputs[1].decrement_use();
                 deleted_indices.push(inputs.len() - 1);
-                op = SupportedOp::Hybrid(crate::circuit::ops::hybrid::HybridOp::GatherElements {
+                op = SupportedOp::Linear(crate::circuit::ops::poly::PolyOp::GatherElements {
                     dim: axis,
                     constant_idx: Some(c.raw_values.map(|x| x as usize)),
                 })
@@ -736,7 +734,7 @@ pub fn new_op_from_onnx(
             SupportedOp::Hybrid(HybridOp::Recip {
                 input_scale: (scale_to_multiplier(in_scale) as f32).into(),
                 output_scale: (scale_to_multiplier(max_scale) as f32).into(),
-                use_range_check_for_int: false,
+                use_range_check_for_int: true,
             })
         }
 

src/graph/vars.rs

@@ -1,4 +1,5 @@
 use std::error::Error;
+use std::fmt::Display;
 
 use crate::tensor::TensorType;
 use crate::tensor::{ValTensor, VarTensor};
@@ -14,6 +15,7 @@ use pyo3::{
 };
 
 use serde::{Deserialize, Serialize};
+use tosubcommand::ToFlags;
 
 use super::*;
 
@@ -40,6 +42,33 @@ pub enum Visibility {
     Fixed,
 }
 
+impl Display for Visibility {
+    fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
+        match self {
+            Visibility::KZGCommit => write!(f, "kzgcommit"),
+            Visibility::Private => write!(f, "private"),
+            Visibility::Public => write!(f, "public"),
+            Visibility::Fixed => write!(f, "fixed"),
+            Visibility::Hashed {
+                hash_is_public,
+                outlets,
+            } => {
+                if *hash_is_public {
+                    write!(f, "hashed/public")
+                } else {
+                    write!(f, "hashed/private/{}", outlets.iter().join(","))
+                }
+            }
+        }
+    }
+}
+
+impl ToFlags for Visibility {
+    fn to_flags(&self) -> Vec<String> {
+        vec![format!("{}", self)]
+    }
+}
+
 impl<'a> From<&'a str> for Visibility {
     fn from(s: &'a str) -> Self {
         if s.contains("hashed/private") {
@@ -202,17 +231,6 @@ impl Visibility {
         vec![]
     }
 }
-impl std::fmt::Display for Visibility {
-    fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
-        match self {
-            Visibility::KZGCommit => write!(f, "kzgcommit"),
-            Visibility::Private => write!(f, "private"),
-            Visibility::Public => write!(f, "public"),
-            Visibility::Fixed => write!(f, "fixed"),
-            Visibility::Hashed { .. } => write!(f, "hashed"),
-        }
-    }
-}
 
 /// Represents the scale of the model input, model parameters.
 #[derive(Clone, Debug, Default, Deserialize, Serialize, PartialEq, PartialOrd)]
@@ -410,7 +428,7 @@ impl<F: PrimeField + TensorType + PartialOrd> ModelVars<F> {
         num_constants: usize,
         module_requires_fixed: bool,
     ) -> Self {
-        info!("number of blinding factors: {}", cs.blinding_factors());
+        debug!("number of blinding factors: {}", cs.blinding_factors());
 
         let advices = (0..3)
             .map(|_| VarTensor::new_advice(cs, logrows, num_inner_cols, var_len))

src/lib.rs

@@ -7,7 +7,6 @@
     overflowing_literals,
     path_statements,
     patterns_in_fns_without_body,
-    private_in_public,
     unconditional_recursion,
     unused,
     unused_allocation,
@@ -33,6 +32,7 @@ use circuit::{table::Range, CheckMode, Tolerance};
 use clap::Args;
 use graph::Visibility;
 use serde::{Deserialize, Serialize};
+use tosubcommand::ToFlags;
 
 /// Methods for configuring tensor operations and assigning values to them in a Halo2 circuit.
 pub mod circuit;
@@ -71,11 +71,34 @@ pub mod tensor;
 #[cfg(all(target_arch = "wasm32", target_os = "unknown"))]
 pub mod wasm;
 
+#[cfg(not(target_arch = "wasm32"))]
+use lazy_static::lazy_static;
+
 /// The denominator in the fixed point representation used when quantizing inputs
 pub type Scale = i32;
 
+#[cfg(not(target_arch = "wasm32"))]
+// Buf writer capacity
+lazy_static! {
+    /// The capacity of the buffer used for writing to disk
+    pub static ref EZKL_BUF_CAPACITY: usize = std::env::var("EZKL_BUF_CAPACITY")
+        .unwrap_or("8000".to_string())
+        .parse()
+        .unwrap();
+
+    /// The serialization format for the keys
+    pub static ref EZKL_KEY_FORMAT: String = std::env::var("EZKL_KEY_FORMAT")
+        .unwrap_or("raw-bytes".to_string());
+}
+
+#[cfg(target_arch = "wasm32")]
+const EZKL_KEY_FORMAT: &str = "raw-bytes";
+
+#[cfg(target_arch = "wasm32")]
+const EZKL_BUF_CAPACITY: &usize = &8000;
+
 /// Parameters specific to a proving run
-#[derive(Debug, Args, Deserialize, Serialize, Clone, PartialEq, PartialOrd)]
+#[derive(Debug, Args, Deserialize, Serialize, Clone, PartialEq, PartialOrd, ToFlags)]
 pub struct RunArgs {
     /// The tolerance for error on model outputs
     #[arg(short = 'T', long, default_value = "0")]
@@ -90,7 +113,7 @@ pub struct RunArgs {
     #[arg(long, default_value = "1")]
     pub scale_rebase_multiplier: u32,
     /// The min and max elements in the lookup table input column
-    #[arg(short = 'B', long, value_parser = parse_tuple::<i128>, default_value = "(-32768,32768)")]
+    #[arg(short = 'B', long, value_parser = parse_key_val::<i128, i128>, default_value = "-32768->32768")]
     pub lookup_range: Range,
     /// The log_2 number of rows
     #[arg(short = 'K', long, default_value = "17")]
@@ -99,7 +122,7 @@ pub struct RunArgs {
     #[arg(short = 'N', long, default_value = "2")]
     pub num_inner_cols: usize,
     /// Hand-written parser for graph variables, eg. batch_size=1
-    #[arg(short = 'V', long, value_parser = parse_key_val::<String, usize>, default_value = "batch_size=1", value_delimiter = ',')]
+    #[arg(short = 'V', long, value_parser = parse_key_val::<String, usize>, default_value = "batch_size->1", value_delimiter = ',')]
     pub variables: Vec<(String, usize)>,
     /// Flags whether inputs are public, private, hashed
     #[arg(long, default_value = "private")]
@@ -157,6 +180,11 @@ impl RunArgs {
         if self.num_inner_cols < 1 {
             return Err("num_inner_cols must be >= 1".into());
         }
+        if self.tolerance.val > 0.0 {
+            if self.output_visibility != Visibility::Public {
+                return Err("tolerance > 0.0 requires output_visibility to be public".into());
+            }
+        }
         Ok(())
     }
 
@@ -181,34 +209,15 @@ fn parse_key_val<T, U>(
     s: &str,
 ) -> Result<(T, U), Box<dyn std::error::Error + Send + Sync + 'static>>
 where
-    T: std::str::FromStr,
+    T: std::str::FromStr + std::fmt::Debug,
     T::Err: std::error::Error + Send + Sync + 'static,
-    U: std::str::FromStr,
+    U: std::str::FromStr + std::fmt::Debug,
     U::Err: std::error::Error + Send + Sync + 'static,
 {
     let pos = s
-        .find('=')
-        .ok_or_else(|| format!("invalid KEY=value: no `=` found in `{s}`"))?;
-    Ok((s[..pos].parse()?, s[pos + 1..].parse()?))
-}
-
-/// Parse a tuple
-fn parse_tuple<T>(s: &str) -> Result<(T, T), Box<dyn std::error::Error + Send + Sync + 'static>>
-where
-    T: std::str::FromStr + Clone,
-    T::Err: std::error::Error + Send + Sync + 'static,
-{
-    let res = s.trim_matches(|p| p == '(' || p == ')').split(',');
-
-    let res = res
-        .map(|x| {
-            // remove blank space
-            let x = x.trim();
-            x.parse::<T>()
-        })
-        .collect::<Result<Vec<_>, _>>()?;
-    if res.len() != 2 {
-        return Err("invalid tuple".into());
-    }
-    Ok((res[0].clone(), res[1].clone()))
+        .find("->")
+        .ok_or_else(|| format!("invalid x->y: no `->` found in `{s}`"))?;
+    let a = s[..pos].parse()?;
+    let b = s[pos + 2..].parse()?;
+    Ok((a, b))
 }

src/pfsys/mod.rs

@@ -8,6 +8,7 @@ use crate::circuit::CheckMode;
 use crate::graph::GraphWitness;
 use crate::pfsys::evm::aggregation::PoseidonTranscript;
 use crate::tensor::TensorType;
+use crate::{EZKL_BUF_CAPACITY, EZKL_KEY_FORMAT};
 use clap::ValueEnum;
 use halo2_proofs::circuit::Value;
 use halo2_proofs::plonk::{
@@ -39,9 +40,19 @@ use std::io::{self, BufReader, BufWriter, Cursor, Write};
 use std::ops::Deref;
 use std::path::PathBuf;
 use thiserror::Error as thisError;
+use tosubcommand::ToFlags;
 
 use halo2curves::bn256::{Bn256, Fr, G1Affine};
 
+fn serde_format_from_str(s: &str) -> halo2_proofs::SerdeFormat {
+    match s {
+        "processed" => halo2_proofs::SerdeFormat::Processed,
+        "raw-bytes-unchecked" => halo2_proofs::SerdeFormat::RawBytesUnchecked,
+        "raw-bytes" => halo2_proofs::SerdeFormat::RawBytes,
+        _ => panic!("invalid serde format"),
+    }
+}
+
 #[allow(missing_docs)]
 #[derive(
     ValueEnum, Copy, Clone, Default, Debug, PartialEq, Eq, Deserialize, Serialize, PartialOrd,
@@ -52,6 +63,25 @@ pub enum ProofType {
     ForAggr,
 }
 
+impl std::fmt::Display for ProofType {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(
+            f,
+            "{}",
+            match self {
+                ProofType::Single => "single",
+                ProofType::ForAggr => "for-aggr",
+            }
+        )
+    }
+}
+
+impl ToFlags for ProofType {
+    fn to_flags(&self) -> Vec<String> {
+        vec![format!("{}", self)]
+    }
+}
+
 impl From<ProofType> for TranscriptType {
     fn from(val: ProofType) -> Self {
         match val {
@@ -154,6 +184,25 @@ pub enum TranscriptType {
     EVM,
 }
 
+impl std::fmt::Display for TranscriptType {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(
+            f,
+            "{}",
+            match self {
+                TranscriptType::Poseidon => "poseidon",
+                TranscriptType::EVM => "evm",
+            }
+        )
+    }
+}
+
+impl ToFlags for TranscriptType {
+    fn to_flags(&self) -> Vec<String> {
+        vec![format!("{}", self)]
+    }
+}
+
 #[cfg(feature = "python-bindings")]
 impl ToPyObject for TranscriptType {
     fn to_object(&self, py: Python) -> PyObject {
@@ -167,8 +216,8 @@ impl ToPyObject for TranscriptType {
 #[cfg(feature = "python-bindings")]
 ///
 pub fn g1affine_to_pydict(g1affine_dict: &PyDict, g1affine: &G1Affine) {
-    let g1affine_x = field_to_string_montgomery(&g1affine.x);
-    let g1affine_y = field_to_string_montgomery(&g1affine.y);
+    let g1affine_x = field_to_string(&g1affine.x);
+    let g1affine_y = field_to_string(&g1affine.y);
     g1affine_dict.set_item("x", g1affine_x).unwrap();
     g1affine_dict.set_item("y", g1affine_y).unwrap();
 }
@@ -178,23 +227,23 @@ use halo2curves::bn256::G1;
 #[cfg(feature = "python-bindings")]
 ///
 pub fn g1_to_pydict(g1_dict: &PyDict, g1: &G1) {
-    let g1_x = field_to_string_montgomery(&g1.x);
-    let g1_y = field_to_string_montgomery(&g1.y);
-    let g1_z = field_to_string_montgomery(&g1.z);
+    let g1_x = field_to_string(&g1.x);
+    let g1_y = field_to_string(&g1.y);
+    let g1_z = field_to_string(&g1.z);
     g1_dict.set_item("x", g1_x).unwrap();
     g1_dict.set_item("y", g1_y).unwrap();
     g1_dict.set_item("z", g1_z).unwrap();
 }
 
-/// converts fp into `Vec<u64>` in Montgomery form
-pub fn field_to_string_montgomery<F: PrimeField + SerdeObject + Serialize>(fp: &F) -> String {
+/// converts fp into a little endian Hex string
+pub fn field_to_string<F: PrimeField + SerdeObject + Serialize>(fp: &F) -> String {
     let repr = serde_json::to_string(&fp).unwrap();
     let b: String = serde_json::from_str(&repr).unwrap();
     b
 }
 
-/// converts `Vec<u64>` in Montgomery form into fp
-pub fn string_to_field_montgomery<F: PrimeField + SerdeObject + Serialize + DeserializeOwned>(
+/// converts a little endian Hex string into a field element
+pub fn string_to_field<F: PrimeField + SerdeObject + Serialize + DeserializeOwned>(
     b: &String,
 ) -> F {
     let repr = serde_json::to_string(&b).unwrap();
@@ -259,7 +308,7 @@ where
         let field_elems: Vec<Vec<String>> = self
             .instances
             .iter()
-            .map(|x| x.iter().map(|fp| field_to_string_montgomery(fp)).collect())
+            .map(|x| x.iter().map(|fp| field_to_string(fp)).collect())
             .collect::<Vec<_>>();
         dict.set_item("instances", field_elems).unwrap();
         let hex_proof = hex::encode(&self.proof);
@@ -315,7 +364,7 @@ where
     /// Saves the Proof to a specified `proof_path`.
     pub fn save(&self, proof_path: &PathBuf) -> Result<(), Box<dyn Error>> {
         let file = std::fs::File::create(proof_path)?;
-        let mut writer = BufWriter::new(file);
+        let mut writer = BufWriter::with_capacity(*EZKL_BUF_CAPACITY, file);
         serde_json::to_writer(&mut writer, &self)?;
         Ok(())
     }
@@ -328,8 +377,10 @@ where
         <C as CurveAffine>::ScalarExt: FromUniformBytes<64>,
     {
         trace!("reading proof");
-        let data = std::fs::read_to_string(proof_path)?;
-        serde_json::from_str(&data).map_err(|e| e.into())
+        let file = std::fs::File::open(proof_path)?;
+        let reader = BufReader::with_capacity(*EZKL_BUF_CAPACITY, file);
+        let proof: Self = serde_json::from_reader(reader)?;
+        Ok(proof)
     }
 }
 
@@ -555,6 +606,7 @@ where
             verifier_params,
             pk.get_vk(),
             strategy,
+            verifier_params.n(),
         )?;
     }
     let elapsed = now.elapsed();
@@ -642,6 +694,7 @@ pub fn verify_proof_circuit<
     params: &'params Scheme::ParamsVerifier,
     vk: &VerifyingKey<Scheme::Curve>,
     strategy: Strategy,
+    orig_n: u64,
 ) -> Result<Strategy::Output, halo2_proofs::plonk::Error>
 where
     Scheme::Scalar: SerdeObject
@@ -662,7 +715,7 @@ where
     trace!("instances {:?}", instances);
 
     let mut transcript = TranscriptReadBuffer::init(Cursor::new(snark.proof.clone()));
-    verify_proof::<Scheme, V, _, TR, _>(params, vk, strategy, instances, &mut transcript)
+    verify_proof::<Scheme, V, _, TR, _>(params, vk, strategy, instances, &mut transcript, orig_n)
 }
 
 /// Loads a [VerifyingKey] at `path`.
@@ -678,13 +731,14 @@ where
     info!("loading verification key from {:?}", path);
     let f =
         File::open(path.clone()).map_err(|_| format!("failed to load vk at {}", path.display()))?;
-    let mut reader = BufReader::new(f);
-    VerifyingKey::<Scheme::Curve>::read::<_, C>(
+    let mut reader = BufReader::with_capacity(*EZKL_BUF_CAPACITY, f);
+    let vk = VerifyingKey::<Scheme::Curve>::read::<_, C>(
         &mut reader,
-        halo2_proofs::SerdeFormat::RawBytes,
+        serde_format_from_str(&EZKL_KEY_FORMAT),
         params,
-    )
-    .map_err(Box::<dyn Error>::from)
+    )?;
+    info!("done loading verification key ✅");
+    Ok(vk)
 }
 
 /// Loads a [ProvingKey] at `path`.
@@ -700,19 +754,20 @@ where
     info!("loading proving key from {:?}", path);
     let f =
         File::open(path.clone()).map_err(|_| format!("failed to load pk at {}", path.display()))?;
-    let mut reader = BufReader::new(f);
-    ProvingKey::<Scheme::Curve>::read::<_, C>(
+    let mut reader = BufReader::with_capacity(*EZKL_BUF_CAPACITY, f);
+    let pk = ProvingKey::<Scheme::Curve>::read::<_, C>(
         &mut reader,
-        halo2_proofs::SerdeFormat::RawBytes,
+        serde_format_from_str(&EZKL_KEY_FORMAT),
         params,
-    )
-    .map_err(Box::<dyn Error>::from)
+    )?;
+    info!("done loading proving key ✅");
+    Ok(pk)
 }
 
 /// Saves a [ProvingKey] to `path`.
 pub fn save_pk<Scheme: CommitmentScheme>(
     path: &PathBuf,
-    vk: &ProvingKey<Scheme::Curve>,
+    pk: &ProvingKey<Scheme::Curve>,
 ) -> Result<(), io::Error>
 where
     Scheme::Curve: SerdeObject + CurveAffine,
@@ -720,9 +775,10 @@ where
 {
     info!("saving proving key 💾");
     let f = File::create(path)?;
-    let mut writer = BufWriter::new(f);
-    vk.write(&mut writer, halo2_proofs::SerdeFormat::RawBytes)?;
+    let mut writer = BufWriter::with_capacity(*EZKL_BUF_CAPACITY, f);
+    pk.write(&mut writer, serde_format_from_str(&EZKL_KEY_FORMAT))?;
     writer.flush()?;
+    info!("done saving proving key ✅");
     Ok(())
 }
 
@@ -737,9 +793,10 @@ where
 {
     info!("saving verification key 💾");
     let f = File::create(path)?;
-    let mut writer = BufWriter::new(f);
-    vk.write(&mut writer, halo2_proofs::SerdeFormat::RawBytes)?;
+    let mut writer = BufWriter::with_capacity(*EZKL_BUF_CAPACITY, f);
+    vk.write(&mut writer, serde_format_from_str(&EZKL_KEY_FORMAT))?;
     writer.flush()?;
+    info!("done saving verification key ✅");
     Ok(())
 }
 
@@ -750,7 +807,7 @@ pub fn save_params<Scheme: CommitmentScheme>(
 ) -> Result<(), io::Error> {
     info!("saving parameters 💾");
     let f = File::create(path)?;
-    let mut writer = BufWriter::new(f);
+    let mut writer = BufWriter::with_capacity(*EZKL_BUF_CAPACITY, f);
     params.write(&mut writer)?;
     writer.flush()?;
     Ok(())
@@ -840,6 +897,7 @@ pub(crate) fn verify_proof_circuit_kzg<
     proof: Snark<Fr, G1Affine>,
     vk: &VerifyingKey<G1Affine>,
     strategy: Strategy,
+    orig_n: u64,
 ) -> Result<Strategy::Output, halo2_proofs::plonk::Error> {
     match proof.transcript_type {
         TranscriptType::EVM => verify_proof_circuit::<
@@ -849,7 +907,7 @@ pub(crate) fn verify_proof_circuit_kzg<
             _,
             _,
             EvmTranscript<G1Affine, _, _, _>,
-        >(&proof, params, vk, strategy),
+        >(&proof, params, vk, strategy, orig_n),
         TranscriptType::Poseidon => verify_proof_circuit::<
             Fr,
             VerifierSHPLONK<'_, Bn256>,
@@ -857,7 +915,7 @@ pub(crate) fn verify_proof_circuit_kzg<
             _,
             _,
             PoseidonTranscript<NativeLoader, _>,
-        >(&proof, params, vk, strategy),
+        >(&proof, params, vk, strategy, orig_n),
     }
 }
 

src/python.rs

@@ -15,10 +15,9 @@ use crate::graph::{
 use crate::pfsys::evm::aggregation::AggregationCircuit;
 use crate::pfsys::{
     load_pk, load_vk, save_params, save_vk, srs::gen_srs as ezkl_gen_srs, srs::load_srs, ProofType,
-    Snark, TranscriptType,
+    TranscriptType,
 };
 use crate::RunArgs;
-use ethers::types::H160;
 use halo2_proofs::poly::kzg::commitment::KZGCommitmentScheme;
 use halo2curves::bn256::{Bn256, Fq, Fr, G1Affine, G1};
 use pyo3::exceptions::{PyIOError, PyRuntimeError};
@@ -26,7 +25,6 @@ use pyo3::prelude::*;
 use pyo3::wrap_pyfunction;
 use pyo3_log;
 use snark_verifier::util::arithmetic::PrimeField;
-use std::str::FromStr;
 use std::{fs::File, path::PathBuf};
 use tokio::runtime::Runtime;
 
@@ -65,9 +63,9 @@ struct PyG1 {
 impl From<G1> for PyG1 {
     fn from(g1: G1) -> Self {
         PyG1 {
-            x: crate::pfsys::field_to_string_montgomery::<Fq>(&g1.x),
-            y: crate::pfsys::field_to_string_montgomery::<Fq>(&g1.y),
-            z: crate::pfsys::field_to_string_montgomery::<Fq>(&g1.z),
+            x: crate::pfsys::field_to_string::<Fq>(&g1.x),
+            y: crate::pfsys::field_to_string::<Fq>(&g1.y),
+            z: crate::pfsys::field_to_string::<Fq>(&g1.z),
         }
     }
 }
@@ -75,9 +73,9 @@ impl From<G1> for PyG1 {
 impl From<PyG1> for G1 {
     fn from(val: PyG1) -> Self {
         G1 {
-            x: crate::pfsys::string_to_field_montgomery::<Fq>(&val.x),
-            y: crate::pfsys::string_to_field_montgomery::<Fq>(&val.y),
-            z: crate::pfsys::string_to_field_montgomery::<Fq>(&val.z),
+            x: crate::pfsys::string_to_field::<Fq>(&val.x),
+            y: crate::pfsys::string_to_field::<Fq>(&val.y),
+            z: crate::pfsys::string_to_field::<Fq>(&val.z),
         }
     }
 }
@@ -108,8 +106,8 @@ pub struct PyG1Affine {
 impl From<G1Affine> for PyG1Affine {
     fn from(g1: G1Affine) -> Self {
         PyG1Affine {
-            x: crate::pfsys::field_to_string_montgomery::<Fq>(&g1.x),
-            y: crate::pfsys::field_to_string_montgomery::<Fq>(&g1.y),
+            x: crate::pfsys::field_to_string::<Fq>(&g1.x),
+            y: crate::pfsys::field_to_string::<Fq>(&g1.y),
         }
     }
 }
@@ -117,8 +115,8 @@ impl From<G1Affine> for PyG1Affine {
 impl From<PyG1Affine> for G1Affine {
     fn from(val: PyG1Affine) -> Self {
         G1Affine {
-            x: crate::pfsys::string_to_field_montgomery::<Fq>(&val.x),
-            y: crate::pfsys::string_to_field_montgomery::<Fq>(&val.y),
+            x: crate::pfsys::string_to_field::<Fq>(&val.x),
+            y: crate::pfsys::string_to_field::<Fq>(&val.y),
         }
     }
 }
@@ -219,53 +217,51 @@ impl Into<PyRunArgs> for RunArgs {
     }
 }
 
-/// Converts 4 u64s to a field element
+/// Converts a felt to big endian
 #[pyfunction(signature = (
-    array,
+    felt,
 ))]
-fn string_to_felt(array: PyFelt) -> PyResult<String> {
-    Ok(format!(
-        "{:?}",
-        crate::pfsys::string_to_field_montgomery::<Fr>(&array)
-    ))
+fn felt_to_big_endian(felt: PyFelt) -> PyResult<String> {
+    let felt = crate::pfsys::string_to_field::<Fr>(&felt);
+    Ok(format!("{:?}", felt))
 }
 
-/// Converts 4 u64s representing a field element directly to an integer
+/// Converts a field element hex string to an integer
 #[pyfunction(signature = (
     array,
 ))]
-fn string_to_int(array: PyFelt) -> PyResult<i128> {
-    let felt = crate::pfsys::string_to_field_montgomery::<Fr>(&array);
+fn felt_to_int(array: PyFelt) -> PyResult<i128> {
+    let felt = crate::pfsys::string_to_field::<Fr>(&array);
     let int_rep = felt_to_i128(felt);
     Ok(int_rep)
 }
 
-/// Converts 4 u64s representing a field element directly to a (rescaled from fixed point scaling) floating point
+/// Converts a field eleement hex string to a floating point number
 #[pyfunction(signature = (
     array,
     scale
 ))]
-fn string_to_float(array: PyFelt, scale: crate::Scale) -> PyResult<f64> {
-    let felt = crate::pfsys::string_to_field_montgomery::<Fr>(&array);
+fn felt_to_float(array: PyFelt, scale: crate::Scale) -> PyResult<f64> {
+    let felt = crate::pfsys::string_to_field::<Fr>(&array);
     let int_rep = felt_to_i128(felt);
     let multiplier = scale_to_multiplier(scale);
     let float_rep = int_rep as f64 / multiplier;
     Ok(float_rep)
 }
 
-/// Converts a floating point element to 4 u64s representing a fixed point field element
+/// Converts a floating point element to a field element hex string
 #[pyfunction(signature = (
 input,
 scale
 ))]
-fn float_to_string(input: f64, scale: crate::Scale) -> PyResult<PyFelt> {
+fn float_to_felt(input: f64, scale: crate::Scale) -> PyResult<PyFelt> {
     let int_rep = quantize_float(&input, 0.0, scale)
         .map_err(|_| PyIOError::new_err("Failed to quantize input"))?;
     let felt = i128_to_felt(int_rep);
-    Ok(crate::pfsys::field_to_string_montgomery::<Fr>(&felt))
+    Ok(crate::pfsys::field_to_string::<Fr>(&felt))
 }
 
-/// Converts a buffer to vector of 4 u64s representing a fixed point field element
+/// Converts a buffer to vector of field elements
 #[pyfunction(signature = (
     buffer
     ))]
@@ -318,7 +314,10 @@ fn buffer_to_felts(buffer: Vec<u8>) -> PyResult<Vec<String>> {
         .map(|x| PrimeField::from_u128(u8_array_to_u128_le(*x)))
         .collect();
 
-    let field_elements: Vec<String> = field_elements.iter().map(|x| format!("{:?}", x)).collect();
+    let field_elements: Vec<String> = field_elements
+        .iter()
+        .map(|x| crate::pfsys::field_to_string::<Fr>(x))
+        .collect();
 
     Ok(field_elements)
 }
@@ -330,7 +329,7 @@ fn buffer_to_felts(buffer: Vec<u8>) -> PyResult<Vec<String>> {
 fn poseidon_hash(message: Vec<PyFelt>) -> PyResult<Vec<PyFelt>> {
     let message: Vec<Fr> = message
         .iter()
-        .map(crate::pfsys::string_to_field_montgomery::<Fr>)
+        .map(crate::pfsys::string_to_field::<Fr>)
         .collect::<Vec<_>>();
 
     let output =
@@ -341,7 +340,7 @@ fn poseidon_hash(message: Vec<PyFelt>) -> PyResult<Vec<PyFelt>> {
 
     let hash = output[0]
         .iter()
-        .map(crate::pfsys::field_to_string_montgomery::<Fr>)
+        .map(crate::pfsys::field_to_string::<Fr>)
         .collect::<Vec<_>>();
     Ok(hash)
 }
@@ -361,7 +360,7 @@ fn kzg_commit(
 ) -> PyResult<Vec<PyG1Affine>> {
     let message: Vec<Fr> = message
         .iter()
-        .map(crate::pfsys::string_to_field_montgomery::<Fr>)
+        .map(crate::pfsys::string_to_field::<Fr>)
         .collect::<Vec<_>>();
 
     let settings = GraphSettings::load(&settings_path)
@@ -525,7 +524,7 @@ fn gen_settings(
     scales = None,
     scale_rebase_multiplier = DEFAULT_SCALE_REBASE_MULTIPLIERS.split(",").map(|x| x.parse().unwrap()).collect(),
     max_logrows = None,
-    div_rebasing = None,
+    only_range_check_rebase = DEFAULT_ONLY_RANGE_CHECK_REBASE.parse().unwrap(),
 ))]
 fn calibrate_settings(
     data: PathBuf,
@@ -536,7 +535,7 @@ fn calibrate_settings(
     scales: Option<Vec<crate::Scale>>,
     scale_rebase_multiplier: Vec<u32>,
     max_logrows: Option<u32>,
-    div_rebasing: Option<bool>,
+    only_range_check_rebase: bool,
 ) -> Result<bool, PyErr> {
     crate::execute::calibrate(
         model,
@@ -546,7 +545,7 @@ fn calibrate_settings(
         lookup_safety_margin,
         scales,
         scale_rebase_multiplier,
-        div_rebasing,
+        only_range_check_rebase,
         max_logrows,
     )
     .map_err(|e| {
@@ -689,14 +688,23 @@ fn prove(
     settings_path=PathBuf::from(DEFAULT_SETTINGS),
     vk_path=PathBuf::from(DEFAULT_VK),
     srs_path=None,
+    non_reduced_srs=DEFAULT_USE_REDUCED_SRS_FOR_VERIFICATION.parse::<bool>().unwrap(),
 ))]
 fn verify(
     proof_path: PathBuf,
     settings_path: PathBuf,
     vk_path: PathBuf,
     srs_path: Option<PathBuf>,
+    non_reduced_srs: bool,
 ) -> Result<bool, PyErr> {
-    crate::execute::verify(proof_path, settings_path, vk_path, srs_path).map_err(|e| {
+    crate::execute::verify(
+        proof_path,
+        settings_path,
+        vk_path,
+        srs_path,
+        non_reduced_srs,
+    )
+    .map_err(|e| {
         let err_str = format!("Failed to run verify: {}", e);
         PyRuntimeError::new_err(err_str)
     })?;
@@ -1022,24 +1030,15 @@ fn verify_evm(
     addr_da: Option<&str>,
     addr_vk: Option<&str>,
 ) -> Result<bool, PyErr> {
-    let addr_verifier = H160::from_str(addr_verifier).map_err(|e| {
-        let err_str = format!("address is invalid: {}", e);
-        PyRuntimeError::new_err(err_str)
-    })?;
+    let addr_verifier = H160Flag::from(addr_verifier);
     let addr_da = if let Some(addr_da) = addr_da {
-        let addr_da = H160::from_str(addr_da).map_err(|e| {
-            let err_str = format!("address is invalid: {}", e);
-            PyRuntimeError::new_err(err_str)
-        })?;
+        let addr_da = H160Flag::from(addr_da);
         Some(addr_da)
     } else {
         None
     };
     let addr_vk = if let Some(addr_vk) = addr_vk {
-        let addr_vk = H160::from_str(addr_vk).map_err(|e| {
-            let err_str = format!("address is invalid: {}", e);
-            PyRuntimeError::new_err(err_str)
-        })?;
+        let addr_vk = H160Flag::from(addr_vk);
         Some(addr_vk)
     } else {
         None
@@ -1097,16 +1096,6 @@ fn create_evm_verifier_aggr(
     Ok(true)
 }
 
-/// print hex representation of a proof
-#[pyfunction(signature = (proof_path))]
-fn print_proof_hex(proof_path: PathBuf) -> Result<String, PyErr> {
-    let proof = Snark::load::<KZGCommitmentScheme<Bn256>>(&proof_path)
-        .map_err(|_| PyIOError::new_err("Failed to load proof"))?;
-
-    let hex_str = hex::encode(proof.proof);
-    Ok(format!("0x{}", hex_str))
-}
-
 // Python Module
 #[pymodule]
 fn ezkl(_py: Python<'_>, m: &PyModule) -> PyResult<()> {
@@ -1115,13 +1104,13 @@ fn ezkl(_py: Python<'_>, m: &PyModule) -> PyResult<()> {
     m.add_class::<PyG1Affine>()?;
     m.add_class::<PyG1>()?;
     m.add_class::<PyTestDataSource>()?;
-    m.add_function(wrap_pyfunction!(string_to_felt, m)?)?;
-    m.add_function(wrap_pyfunction!(string_to_int, m)?)?;
-    m.add_function(wrap_pyfunction!(string_to_float, m)?)?;
+    m.add_function(wrap_pyfunction!(felt_to_big_endian, m)?)?;
+    m.add_function(wrap_pyfunction!(felt_to_int, m)?)?;
+    m.add_function(wrap_pyfunction!(felt_to_float, m)?)?;
     m.add_function(wrap_pyfunction!(kzg_commit, m)?)?;
     m.add_function(wrap_pyfunction!(swap_proof_commitments, m)?)?;
     m.add_function(wrap_pyfunction!(poseidon_hash, m)?)?;
-    m.add_function(wrap_pyfunction!(float_to_string, m)?)?;
+    m.add_function(wrap_pyfunction!(float_to_felt, m)?)?;
     m.add_function(wrap_pyfunction!(buffer_to_felts, m)?)?;
     m.add_function(wrap_pyfunction!(gen_vk_from_pk_aggr, m)?)?;
     m.add_function(wrap_pyfunction!(gen_vk_from_pk_single, m)?)?;
@@ -1145,7 +1134,6 @@ fn ezkl(_py: Python<'_>, m: &PyModule) -> PyResult<()> {
     m.add_function(wrap_pyfunction!(deploy_vk_evm, m)?)?;
     m.add_function(wrap_pyfunction!(deploy_da_evm, m)?)?;
     m.add_function(wrap_pyfunction!(verify_evm, m)?)?;
-    m.add_function(wrap_pyfunction!(print_proof_hex, m)?)?;
     m.add_function(wrap_pyfunction!(setup_test_evm_witness, m)?)?;
     m.add_function(wrap_pyfunction!(create_evm_verifier_aggr, m)?)?;
     m.add_function(wrap_pyfunction!(create_evm_data_attestation, m)?)?;

src/tensor/mod.rs

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

src/tensor/ops.rs

@@ -243,7 +243,7 @@ pub fn and<
 /// Some(&[1, 0, 1, 0, 1, 0]),
 /// &[2, 3],
 /// ).unwrap();
-/// let result = equals(&a, &b).unwrap().0;
+/// let result = equals(&a, &b).unwrap();
 /// let expected = Tensor::<i128>::new(Some(&[1, 0, 1, 0, 1, 1]), &[2, 3]).unwrap();
 /// assert_eq!(result, expected);
 /// ```
@@ -260,7 +260,7 @@ pub fn equals<
 >(
     a: &Tensor<T>,
     b: &Tensor<T>,
-) -> Result<(Tensor<T>, Vec<Tensor<T>>), TensorError> {
+) -> Result<Tensor<T>, TensorError> {
     let a = a.clone();
     let b = b.clone();
 
@@ -268,7 +268,7 @@ pub fn equals<
 
     let result = nonlinearities::kronecker_delta(&diff);
 
-    Ok((result, vec![diff]))
+    Ok(result)
 }
 
 /// Greater than operation.
@@ -289,7 +289,7 @@ pub fn equals<
 /// ).unwrap();
 /// let result = greater(&a, &b).unwrap();
 /// let expected = Tensor::<i128>::new(Some(&[0, 1, 1, 0, 0, 0]), &[2, 3]).unwrap();
-/// assert_eq!(result.0, expected);
+/// assert_eq!(result, expected);
 /// ```
 pub fn greater<
     T: TensorType
@@ -302,7 +302,7 @@ pub fn greater<
 >(
     a: &Tensor<T>,
     b: &Tensor<T>,
-) -> Result<(Tensor<T>, Vec<Tensor<T>>), TensorError> {
+) -> Result<Tensor<T>, TensorError> {
     let mask_inter = (a.clone() - b.clone())?;
     let mask = mask_inter.map(|x| {
         if x > T::zero().ok_or(TensorError::Unsupported).unwrap() {
@@ -311,7 +311,7 @@ pub fn greater<
             T::zero().ok_or(TensorError::Unsupported).unwrap()
         }
     });
-    Ok((mask, vec![mask_inter]))
+    Ok(mask)
 }
 
 /// Greater equals than operation.
@@ -332,7 +332,7 @@ pub fn greater<
 /// ).unwrap();
 /// let result = greater_equal(&a, &b).unwrap();
 /// let expected = Tensor::<i128>::new(Some(&[1, 1, 1, 1, 0, 0]), &[2, 3]).unwrap();
-/// assert_eq!(result.0, expected);
+/// assert_eq!(result, expected);
 /// ```
 pub fn greater_equal<
     T: TensorType
@@ -345,7 +345,7 @@ pub fn greater_equal<
 >(
     a: &Tensor<T>,
     b: &Tensor<T>,
-) -> Result<(Tensor<T>, Vec<Tensor<T>>), TensorError> {
+) -> Result<Tensor<T>, TensorError> {
     let mask_inter = (a.clone() - b.clone())?;
     let mask = mask_inter.map(|x| {
         if x >= T::zero().ok_or(TensorError::Unsupported).unwrap() {
@@ -354,7 +354,7 @@ pub fn greater_equal<
             T::zero().ok_or(TensorError::Unsupported).unwrap()
         }
     });
-    Ok((mask, vec![mask_inter]))
+    Ok(mask)
 }
 
 /// Less than to operation.
@@ -375,7 +375,7 @@ pub fn greater_equal<
 /// ).unwrap();
 /// let result = less(&a, &b).unwrap();
 /// let expected = Tensor::<i128>::new(Some(&[0, 1, 0, 0, 0, 1]), &[2, 3]).unwrap();
-/// assert_eq!(result.0, expected);
+/// assert_eq!(result, expected);
 /// ```
 ///
 pub fn less<
@@ -389,7 +389,7 @@ pub fn less<
 >(
     a: &Tensor<T>,
     b: &Tensor<T>,
-) -> Result<(Tensor<T>, Vec<Tensor<T>>), TensorError> {
+) -> Result<Tensor<T>, TensorError> {
     // a < b <=> b > a
     greater(b, a)
 }
@@ -412,7 +412,7 @@ pub fn less<
 /// ).unwrap();
 /// let result = less_equal(&a, &b).unwrap();
 /// let expected = Tensor::<i128>::new(Some(&[1, 1, 0, 1, 1, 1]), &[2, 3]).unwrap();
-/// assert_eq!(result.0, expected);
+/// assert_eq!(result, expected);
 /// ```
 ///
 pub fn less_equal<
@@ -426,7 +426,7 @@ pub fn less_equal<
 >(
     a: &Tensor<T>,
     b: &Tensor<T>,
-) -> Result<(Tensor<T>, Vec<Tensor<T>>), TensorError> {
+) -> Result<Tensor<T>, TensorError> {
     // a < b <=> b > a
     greater_equal(b, a)
 }
@@ -2300,12 +2300,12 @@ pub fn deconv<
 ///     Some(&[5, 2, 3, 0, 4, -1, 3, 1, 6]),
 ///     &[1, 1, 3, 3],
 /// ).unwrap();
-/// let pooled = sumpool(&x, [(0, 0); 2], (1, 1), (2, 2), false).unwrap().0;
+/// let pooled = sumpool(&x, [(0, 0); 2], (1, 1), (2, 2), false).unwrap();
 /// let expected: Tensor<i128> = Tensor::<i128>::new(Some(&[11, 8, 8, 10]), &[1, 1, 2, 2]).unwrap();
 /// assert_eq!(pooled, expected);
 ///
 /// // This time with normalization
-/// let pooled = sumpool(&x, [(0, 0); 2], (1, 1), (2, 2), true).unwrap().0;
+/// let pooled = sumpool(&x, [(0, 0); 2], (1, 1), (2, 2), true).unwrap();
 /// let expected: Tensor<i128> = Tensor::<i128>::new(Some(&[3, 2, 2, 3]), &[1, 1, 2, 2]).unwrap();
 /// assert_eq!(pooled, expected);
 /// ```
@@ -2315,7 +2315,7 @@ pub fn sumpool(
     stride: (usize, usize),
     kernel_shape: (usize, usize),
     normalize: bool,
-) -> Result<(Tensor<i128>, Vec<Tensor<i128>>), TensorError> {
+) -> Result<Tensor<i128>, TensorError> {
     let image_dims = image.dims();
     let batch_size = image_dims[0];
     let image_channels = image_dims[1];
@@ -2345,15 +2345,12 @@ pub fn sumpool(
     let mut combined = res.combine()?;
     combined.reshape(&[&[batch_size, image_channels], shape].concat())?;
 
-    let mut inter = vec![];
-
     if normalize {
-        inter.push(combined.clone());
         let norm = kernel.len();
         combined = nonlinearities::const_div(&combined, norm as f64);
     }
 
-    Ok((combined, inter))
+    Ok(combined)
 }
 
 /// Applies 2D max pooling over a 4D tensor of shape B x C x H x W.
@@ -3048,11 +3045,7 @@ pub mod nonlinearities {
     }
 
     /// softmax layout
-    pub fn softmax_axes(
-        a: &Tensor<i128>,
-        scale: f64,
-        axes: &[usize],
-    ) -> (Tensor<i128>, Vec<Tensor<i128>>) {
+    pub fn softmax_axes(a: &Tensor<i128>, scale: f64, axes: &[usize]) -> Tensor<i128> {
         // we want this to be as small as possible so we set the output scale to 1
         let dims = a.dims();
 
@@ -3060,8 +3053,6 @@ pub mod nonlinearities {
             return softmax(a, scale);
         }
 
-        let mut intermediate_values = vec![];
-
         let cartesian_coord = dims[..dims.len() - 1]
             .iter()
             .map(|x| 0..*x)
@@ -3084,8 +3075,7 @@ pub mod nonlinearities {
 
             let res = softmax(&softmax_input, scale);
 
-            outputs.push(res.0);
-            intermediate_values.extend(res.1);
+            outputs.push(res);
         }
 
         let mut res = Tensor::new(Some(&outputs), &[outputs.len()])
@@ -3093,7 +3083,7 @@ pub mod nonlinearities {
             .combine()
             .unwrap();
         res.reshape(dims).unwrap();
-        (res, intermediate_values)
+        res
     }
 
     /// Applies softmax
@@ -3110,24 +3100,20 @@ pub mod nonlinearities {
     ///     Some(&[2, 2, 3, 2, 2, 0]),
     ///     &[2, 3],
     /// ).unwrap();
-    /// let result = softmax(&x, 128.0).0;
+    /// let result = softmax(&x, 128.0);
     /// // doubles the scale of the input
     /// let expected = Tensor::<i128>::new(Some(&[2730, 2730, 2751, 2730, 2730, 2688]), &[2, 3]).unwrap();
     /// assert_eq!(result, expected);
     /// ```
-    pub fn softmax(a: &Tensor<i128>, scale: f64) -> (Tensor<i128>, Vec<Tensor<i128>>) {
+    pub fn softmax(a: &Tensor<i128>, scale: f64) -> Tensor<i128> {
         // the more accurate calculation is commented out and we implement as below so it matches the steps in layout
-        let mut intermediate_values = vec![];
-
-        intermediate_values.push(a.clone());
 
         let exp = exp(a, scale);
 
         let sum = sum(&exp).unwrap();
-        intermediate_values.push(sum.clone());
         let inv_denom = recip(&sum, scale, scale);
 
-        ((exp * inv_denom).unwrap(), intermediate_values)
+        (exp * inv_denom).unwrap()
     }
 
     /// Applies range_check_percent
@@ -3787,6 +3773,30 @@ pub mod nonlinearities {
         .unwrap()
     }
 
+    /// Elementwise inverse.
+    /// # Arguments
+    /// * `out_scale` - Single value
+    /// # Examples
+    /// ```
+    /// use ezkl::tensor::Tensor;
+    /// use ezkl::tensor::ops::nonlinearities::zero_recip;
+    /// let k = 2_f64;
+    /// let result = zero_recip(1.0);
+    /// let expected = Tensor::<i128>::new(Some(&[4503599627370496]), &[1]).unwrap();
+    /// assert_eq!(result, expected);
+    /// ```
+    pub fn zero_recip(out_scale: f64) -> Tensor<i128> {
+        let a = Tensor::<i128>::new(Some(&[0]), &[1]).unwrap();
+
+        a.par_enum_map(|_, a_i| {
+            let rescaled = a_i as f64;
+            let denom = (1_f64) / (rescaled + f64::EPSILON);
+            let d_inv_x = out_scale * denom;
+            Ok::<_, TensorError>(d_inv_x.round() as i128)
+        })
+        .unwrap()
+    }
+
     /// Elementwise greater than
     /// # Arguments
     ///

src/tensor/val.rs

@@ -454,12 +454,12 @@ impl<F: PrimeField + TensorType + PartialOrd> ValTensor<F> {
     }
 
     /// Calls `pad_to_zero_rem` on the inner tensor.
-    pub fn pad_to_zero_rem(&mut self, n: usize) -> Result<(), Box<dyn Error>> {
+    pub fn pad_to_zero_rem(&mut self, n: usize, pad: ValType<F>) -> Result<(), Box<dyn Error>> {
         match self {
             ValTensor::Value {
                 inner: v, dims: d, ..
             } => {
-                *v = v.pad_to_zero_rem(n)?;
+                *v = v.pad_to_zero_rem(n, pad)?;
                 *d = v.dims().to_vec();
             }
             ValTensor::Instance { .. } => {
@@ -672,7 +672,7 @@ impl<F: PrimeField + TensorType + PartialOrd> ValTensor<F> {
                 }
                 Ok(indices)
             }
-            ValTensor::Instance { .. } => Err(TensorError::WrongMethod),
+            ValTensor::Instance { .. } => Ok(vec![]),
         }
     }
 
@@ -690,7 +690,7 @@ impl<F: PrimeField + TensorType + PartialOrd> ValTensor<F> {
                 }
                 Ok(indices)
             }
-            ValTensor::Instance { .. } => Err(TensorError::WrongMethod),
+            ValTensor::Instance { .. } => Ok(vec![]),
         }
     }
 
@@ -709,7 +709,11 @@ impl<F: PrimeField + TensorType + PartialOrd> ValTensor<F> {
                 *d = v.dims().to_vec();
             }
             ValTensor::Instance { .. } => {
-                return Err(TensorError::WrongMethod);
+                if indices.is_empty() {
+                    return Ok(());
+                } else {
+                    return Err(TensorError::WrongMethod);
+                }
             }
         }
         Ok(())
@@ -871,3 +875,30 @@ 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)
+    }
+}

src/tensor/var.rs

@@ -33,10 +33,7 @@ pub enum VarTensor {
 impl VarTensor {
     ///
     pub fn is_advice(&self) -> bool {
-        match self {
-            VarTensor::Advice { .. } => true,
-            _ => false,
-        }
+        matches!(self, VarTensor::Advice { .. })
     }
 
     ///

src/wasm.rs

@@ -69,19 +69,19 @@ pub fn encodeVerifierCalldata(
     Ok(encoded)
 }
 
-/// Converts 4 u64s to a field element
+/// Converts a hex string to a byte array
 #[wasm_bindgen]
 #[allow(non_snake_case)]
-pub fn stringToFelt(array: wasm_bindgen::Clamped<Vec<u8>>) -> Result<String, JsError> {
+pub fn feltToBigEndian(array: wasm_bindgen::Clamped<Vec<u8>>) -> Result<String, JsError> {
     let felt: Fr = serde_json::from_slice(&array[..])
         .map_err(|e| JsError::new(&format!("Failed to deserialize field element: {}", e)))?;
     Ok(format!("{:?}", felt))
 }
 
-/// Converts 4 u64s representing a field element directly to an integer
+/// Converts a hex string to a byte array
 #[wasm_bindgen]
 #[allow(non_snake_case)]
-pub fn stringToInt(
+pub fn feltToInt(
     array: wasm_bindgen::Clamped<Vec<u8>>,
 ) -> Result<wasm_bindgen::Clamped<Vec<u8>>, JsError> {
     let felt: Fr = serde_json::from_slice(&array[..])
@@ -92,10 +92,10 @@ pub fn stringToInt(
     ))
 }
 
-/// Converts 4 u64s representing a field element directly to a (rescaled from fixed point scaling) floating point
+/// Converts felts to a floating point element
 #[wasm_bindgen]
 #[allow(non_snake_case)]
-pub fn stringToFloat(
+pub fn feltToFloat(
     array: wasm_bindgen::Clamped<Vec<u8>>,
     scale: crate::Scale,
 ) -> Result<f64, JsError> {
@@ -106,26 +106,26 @@ pub fn stringToFloat(
     Ok(int_rep as f64 / multiplier)
 }
 
-/// Converts a floating point element to 4 u64s representing a fixed point field element
+/// Converts a floating point number to a hex string representing a fixed point field element
 #[wasm_bindgen]
 #[allow(non_snake_case)]
-pub fn floatTostring(
+pub fn floatToFelt(
     input: f64,
     scale: crate::Scale,
 ) -> Result<wasm_bindgen::Clamped<Vec<u8>>, JsError> {
     let int_rep =
         quantize_float(&input, 0.0, scale).map_err(|e| JsError::new(&format!("{}", e)))?;
     let felt = i128_to_felt(int_rep);
-    let vec = crate::pfsys::field_to_string_montgomery::<halo2curves::bn256::Fr>(&felt);
+    let vec = crate::pfsys::field_to_string::<halo2curves::bn256::Fr>(&felt);
     Ok(wasm_bindgen::Clamped(serde_json::to_vec(&vec).map_err(
-        |e| JsError::new(&format!("Failed to serialize string_montgomery{}", e)),
+        |e| JsError::new(&format!("Failed to serialize a float to felt{}", e)),
     )?))
 }
 
 /// Converts a buffer to vector of 4 u64s representing a fixed point field element
 #[wasm_bindgen]
 #[allow(non_snake_case)]
-pub fn bufferToVecOfstring(
+pub fn bufferToVecOfFelt(
     buffer: wasm_bindgen::Clamped<Vec<u8>>,
 ) -> Result<wasm_bindgen::Clamped<Vec<u8>>, JsError> {
     // Convert the buffer to a slice
@@ -211,7 +211,7 @@ pub fn genWitness(
         .map_err(|e| JsError::new(&format!("{}", e)))?;
 
     let witness = circuit
-        .forward(&mut input, None, None)
+        .forward(&mut input, None, None, false)
         .map_err(|e| JsError::new(&format!("{}", e)))?;
 
     serde_json::to_vec(&witness)
@@ -311,13 +311,19 @@ pub fn verify(
     let vk = VerifyingKey::<G1Affine>::read::<_, GraphCircuit>(
         &mut reader,
         halo2_proofs::SerdeFormat::RawBytes,
-        circuit_settings,
+        circuit_settings.clone(),
     )
     .map_err(|e| JsError::new(&format!("Failed to deserialize vk: {}", e)))?;
 
     let strategy = KZGSingleStrategy::new(params.verifier_params());
 
-    let result = verify_proof_circuit_kzg(params.verifier_params(), snark, &vk, strategy);
+    let result = verify_proof_circuit_kzg(
+        params.verifier_params(),
+        snark,
+        &vk,
+        strategy,
+        1 << circuit_settings.run_args.logrows,
+    );
 
     match result {
         Ok(_) => Ok(true),
@@ -387,15 +393,6 @@ pub fn prove(
         .into_bytes())
 }
 
-/// print hex representation of a proof
-#[wasm_bindgen]
-#[allow(non_snake_case)]
-pub fn printProofHex(proof: wasm_bindgen::Clamped<Vec<u8>>) -> Result<String, JsError> {
-    let proof: crate::pfsys::Snark<Fr, G1Affine> = serde_json::from_slice(&proof[..])
-        .map_err(|e| JsError::new(&format!("Failed to deserialize proof: {}", e)))?;
-    let hex_str = hex::encode(proof.proof);
-    Ok(format!("0x{}", hex_str))
-}
 // VALIDATION FUNCTIONS
 
 /// Witness file validation

tests/integration_tests.rs

@@ -2,6 +2,8 @@
 #[cfg(test)]
 mod native_tests {
 
+    use ezkl::circuit::Tolerance;
+    use ezkl::fieldutils::{felt_to_i128, i128_to_felt};
     // use ezkl::circuit::table::RESERVED_BLINDING_ROWS_PAD;
     use ezkl::graph::input::{FileSource, FileSourceInner, GraphData};
     use ezkl::graph::{DataSource, GraphSettings, GraphWitness};
@@ -192,90 +194,90 @@ mod native_tests {
     ];
 
     const TESTS: [&str; 77] = [
-        "1l_mlp",
+        "1l_mlp", //0
         "1l_slice",
         "1l_concat",
         "1l_flatten",
         // "1l_average",
         "1l_div",
-        "1l_pad",
+        "1l_pad", // 5
         "1l_reshape",
         "1l_eltwise_div",
         "1l_sigmoid",
         "1l_sqrt",
-        "1l_softmax",
+        "1l_softmax", //10
         // "1l_instance_norm",
         "1l_batch_norm",
         "1l_prelu",
         "1l_leakyrelu",
         "1l_gelu_noappx",
         // "1l_gelu_tanh_appx",
-        "1l_relu",
+        "1l_relu", //15
         "1l_downsample",
         "1l_tanh",
         "2l_relu_sigmoid_small",
         "2l_relu_fc",
-        "2l_relu_small",
+        "2l_relu_small", //20
         "2l_relu_sigmoid",
         "1l_conv",
         "2l_sigmoid_small",
         "2l_relu_sigmoid_conv",
-        "3l_relu_conv_fc",
+        "3l_relu_conv_fc", //25
         "4l_relu_conv_fc",
         "1l_erf",
         "1l_var",
-        "1l_elu", //30
-        "min",
+        "1l_elu",
+        "min", //30
         "max",
         "1l_max_pool",
         "1l_conv_transpose",
-        "1l_upsample", //35
-        "1l_identity",
+        "1l_upsample",
+        "1l_identity", //35
         "idolmodel",
         "trig",
         "prelu_gmm",
-        "lstm", //40
-        "rnn",
+        "lstm",
+        "rnn", //40
         "quantize_dequantize",
         "1l_where",
         "boolean",
         "boolean_identity",
-        "decision_tree", // "variable_cnn",
+        "decision_tree", // 45
         "random_forest",
         "gradient_boosted_trees",
         "1l_topk",
-        "xgboost", //50
-        "lightgbm",
+        "xgboost",
+        "lightgbm", //50
         "hummingbird_decision_tree",
         "oh_decision_tree",
         "linear_svc",
         "gather_elements",
-        "less",
+        "less", //55
         "xgboost_reg",
         "1l_powf",
         "scatter_elements",
-        "1l_linear", //60
-        "linear_regression",
+        "1l_linear",
+        "linear_regression", //60
         "sklearn_mlp",
         "1l_mean",
         "rounding_ops",
         // "mean_as_constrain",
         "arange",
-        "layernorm",
+        "layernorm", //65
         "bitwise_ops",
         "blackman_window",
-        "softsign", //70
+        "softsign", //68
         "softplus",
-        "selu",
+        "selu", //70
         "hard_sigmoid",
         "log_softmax",
         "eye",
         "ltsf",
-        "remainder",
+        "remainder", //75
         "bitshift",
     ];
 
-    const WASM_TESTS: [&str; 48] = [
+    const WASM_TESTS: [&str; 46] = [
         "1l_mlp",
         "1l_slice",
         "1l_concat",
@@ -324,8 +326,6 @@ mod native_tests {
         "1l_where",
         "boolean",
         "boolean_identity",
-        "decision_tree", // "variable_cnn",
-        "random_forest",
         "gradient_boosted_trees",
         "1l_topk",
         // "xgboost",
@@ -477,6 +477,7 @@ mod native_tests {
             use crate::native_tests::kzg_fuzz;
             use crate::native_tests::render_circuit;
             use crate::native_tests::model_serialization_different_binaries;
+            use rand::Rng;
             use tempdir::TempDir;
 
             #[test]
@@ -496,7 +497,7 @@ mod native_tests {
                 crate::native_tests::init_binary();
                 let test_dir = TempDir::new(test).unwrap();
                 let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
-                mock(path, test.to_string(), "public", "fixed", "public", 1, "accuracy", None);
+                mock(path, test.to_string(), "public", "fixed", "public", 1, "accuracy", None, 0.0);
                 test_dir.close().unwrap();
             }
         });
@@ -560,7 +561,7 @@ mod native_tests {
                 crate::native_tests::setup_py_env();
                 let test_dir = TempDir::new(test).unwrap();
                 let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
-                accuracy_measurement(path, test.to_string(), "private", "private", "public", 1, "resources", 18.0, false);
+                accuracy_measurement(path, test.to_string(), "private", "private", "public", 1, "resources", 3.1, false);
                 test_dir.close().unwrap();
             }
 
@@ -569,10 +570,23 @@ mod native_tests {
                 crate::native_tests::init_binary();
                 let test_dir = TempDir::new(test).unwrap();
                 let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
-                mock(path, test.to_string(), "private", "private", "public", 1, "resources", None);
+                mock(path, test.to_string(), "private", "private", "public", 1, "resources", None, 0.0);
                 test_dir.close().unwrap();
             }
 
+            #(#[test_case(TESTS[N])])*
+            fn mock_tolerance_public_outputs_(test: &str) {
+                crate::native_tests::init_binary();
+                let test_dir = TempDir::new(test).unwrap();
+                let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
+                // gen random number between 0.0 and 1.0
+                let tolerance = rand::thread_rng().gen_range(0.0..1.0) * 100.0;
+                mock(path, test.to_string(), "private", "private", "public", 1, "resources", None, tolerance);
+                test_dir.close().unwrap();
+            }
+
+
+
             #(#[test_case(TESTS[N])])*
             fn mock_large_batch_public_outputs_(test: &str) {
                 crate::native_tests::init_binary();
@@ -580,7 +594,7 @@ mod native_tests {
                 let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
                 let large_batch_dir = &format!("large_batches_{}", test);
                 crate::native_tests::mk_data_batches_(path, test, &large_batch_dir, 10);
-                mock(path, large_batch_dir.to_string(), "private", "private", "public", 10, "resources", None);
+                mock(path, large_batch_dir.to_string(), "private", "private", "public", 10, "resources", None, 0.0);
                 test_dir.close().unwrap();
             }
 
@@ -589,7 +603,7 @@ mod native_tests {
                 crate::native_tests::init_binary();
                 let test_dir = TempDir::new(test).unwrap();
                 let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
-                mock(path, test.to_string(), "public", "private", "private", 1, "resources", None);
+                mock(path, test.to_string(), "public", "private", "private", 1, "resources", None, 0.0);
                 test_dir.close().unwrap();
             }
 
@@ -598,7 +612,7 @@ mod native_tests {
                 crate::native_tests::init_binary();
                 let test_dir = TempDir::new(test).unwrap();
                 let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
-                mock(path, test.to_string(), "fixed", "private", "private", 1, "resources", None);
+                mock(path, test.to_string(), "fixed", "private", "private", 1, "resources", None, 0.0);
                 test_dir.close().unwrap();
             }
 
@@ -607,7 +621,7 @@ mod native_tests {
                 crate::native_tests::init_binary();
                 let test_dir = TempDir::new(test).unwrap();
                 let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
-                mock(path, test.to_string(), "private", "private", "fixed", 1, "resources", None);
+                mock(path, test.to_string(), "private", "private", "fixed", 1, "resources", None, 0.0);
                 test_dir.close().unwrap();
             }
 
@@ -616,7 +630,7 @@ mod native_tests {
                 crate::native_tests::init_binary();
                 let test_dir = TempDir::new(test).unwrap();
                 let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
-                mock(path, test.to_string(), "private", "fixed", "private", 1, "resources", None);
+                mock(path, test.to_string(), "private", "fixed", "private", 1, "resources", None, 0.0);
                 test_dir.close().unwrap();
             }
 
@@ -625,7 +639,7 @@ mod native_tests {
                 crate::native_tests::init_binary();
                 let test_dir = TempDir::new(test).unwrap();
                 let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
-                mock(path, test.to_string(), "hashed", "private", "public", 1, "resources", None);
+                mock(path, test.to_string(), "hashed", "private", "public", 1, "resources", None, 0.0);
                 test_dir.close().unwrap();
             }
 
@@ -634,7 +648,7 @@ mod native_tests {
                 crate::native_tests::init_binary();
                 let test_dir = TempDir::new(test).unwrap();
                 let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
-                mock(path, test.to_string(), "kzgcommit", "private", "public", 1, "resources", None);
+                mock(path, test.to_string(), "kzgcommit", "private", "public", 1, "resources", None, 0.0);
                 test_dir.close().unwrap();
             }
 
@@ -644,7 +658,7 @@ mod native_tests {
                 crate::native_tests::init_binary();
                 let test_dir = TempDir::new(test).unwrap();
                 let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
-                mock(path, test.to_string(), "private", "hashed", "public", 1, "resources", None);
+                mock(path, test.to_string(), "private", "hashed", "public", 1, "resources", None, 0.0);
                 test_dir.close().unwrap();
             }
 
@@ -654,7 +668,7 @@ mod native_tests {
                 crate::native_tests::init_binary();
                 let test_dir = TempDir::new(test).unwrap();
                 let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
-                mock(path, test.to_string(), "private", "kzgcommit", "public", 1, "resources", None);
+                mock(path, test.to_string(), "private", "kzgcommit", "public", 1, "resources", None, 0.0);
                 test_dir.close().unwrap();
             }
 
@@ -663,7 +677,7 @@ mod native_tests {
                 crate::native_tests::init_binary();
                 let test_dir = TempDir::new(test).unwrap();
                 let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
-                mock(path, test.to_string(), "public", "private", "hashed", 1, "resources", None);
+                mock(path, test.to_string(), "public", "private", "hashed", 1, "resources", None, 0.0);
                 test_dir.close().unwrap();
             }
 
@@ -673,7 +687,7 @@ mod native_tests {
                 crate::native_tests::init_binary();
                 let test_dir = TempDir::new(test).unwrap();
                 let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
-                mock(path, test.to_string(), "public", "private", "kzgcommit", 1, "resources", None);
+                mock(path, test.to_string(), "public", "private", "kzgcommit", 1, "resources", None, 0.0);
                 test_dir.close().unwrap();
             }
 
@@ -682,7 +696,7 @@ mod native_tests {
                 crate::native_tests::init_binary();
                 let test_dir = TempDir::new(test).unwrap();
                 let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
-                mock(path, test.to_string(), "public", "fixed", "hashed", 1, "resources", None);
+                mock(path, test.to_string(), "public", "fixed", "hashed", 1, "resources", None, 0.0);
                 test_dir.close().unwrap();
             }
 
@@ -692,7 +706,7 @@ mod native_tests {
                 crate::native_tests::init_binary();
                 let test_dir = TempDir::new(test).unwrap();
                 let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
-                mock(path, test.to_string(), "public", "kzgcommit", "hashed", 1, "resources", None);
+                mock(path, test.to_string(), "public", "kzgcommit", "hashed", 1, "resources", None, 0.0);
                 test_dir.close().unwrap();
             }
 
@@ -702,7 +716,7 @@ mod native_tests {
                 crate::native_tests::init_binary();
                 let test_dir = TempDir::new(test).unwrap();
                 let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
-                mock(path, test.to_string(), "kzgcommit", "kzgcommit", "kzgcommit", 1, "resources", None);
+                mock(path, test.to_string(), "kzgcommit", "kzgcommit", "kzgcommit", 1, "resources", None, 0.0);
                 test_dir.close().unwrap();
             }
 
@@ -712,7 +726,7 @@ mod native_tests {
                 crate::native_tests::init_binary();
                 let test_dir = TempDir::new(test).unwrap();
                 let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
-                mock(path, test.to_string(), "hashed", "private", "hashed", 1, "resources", None);
+                mock(path, test.to_string(), "hashed", "private", "hashed", 1, "resources", None, 0.0);
                 test_dir.close().unwrap();
             }
 
@@ -722,7 +736,7 @@ mod native_tests {
                 let test_dir = TempDir::new(test).unwrap();
                 let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
                 // needs an extra row for the large model
-                mock(path, test.to_string(),"hashed", "hashed", "public", 1, "resources", None);
+                mock(path, test.to_string(),"hashed", "hashed", "public", 1, "resources", None, 0.0);
                 test_dir.close().unwrap();
             }
 
@@ -732,7 +746,7 @@ mod native_tests {
                 let test_dir = TempDir::new(test).unwrap();
                 let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
                 // needs an extra row for the large model
-                mock(path, test.to_string(),"hashed", "hashed", "hashed", 1, "resources", None);
+                mock(path, test.to_string(),"hashed", "hashed", "hashed", 1, "resources", None, 0.0);
                 test_dir.close().unwrap();
             }
 
@@ -828,7 +842,7 @@ mod native_tests {
 
             });
 
-            seq!(N in 0..=47 {
+            seq!(N in 0..=45 {
 
                 #(#[test_case(WASM_TESTS[N])])*
                 fn kzg_prove_and_verify_with_overflow_(test: &str) {
@@ -876,7 +890,7 @@ mod native_tests {
                 crate::native_tests::init_binary();
                 let test_dir = TempDir::new(test).unwrap();
                 let path = test_dir.path().to_str().unwrap(); crate::native_tests::mv_test_(path, test);
-                mock(path, test.to_string(), "private", "fixed", "public", 1, "resources", None);
+                mock(path, test.to_string(), "private", "fixed", "public", 1, "resources", None, 0.0);
                 test_dir.close().unwrap();
             }
         });
@@ -1273,7 +1287,9 @@ mod native_tests {
         batch_size: usize,
         cal_target: &str,
         scales_to_use: Option<Vec<u32>>,
+        tolerance: f32,
     ) {
+        let mut tolerance = tolerance;
         gen_circuit_settings_and_witness(
             test_dir,
             example_name.clone(),
@@ -1285,19 +1301,131 @@ mod native_tests {
             scales_to_use,
             2,
             false,
+            &mut tolerance,
         );
 
-        let status = Command::new(format!("{}/release/ezkl", *CARGO_TARGET_DIR))
-            .args([
-                "mock",
-                "-W",
-                format!("{}/{}/witness.json", test_dir, example_name).as_str(),
-                "-M",
-                format!("{}/{}/network.compiled", test_dir, example_name).as_str(),
-            ])
-            .status()
-            .expect("failed to execute process");
-        assert!(status.success());
+        if tolerance > 0.0 {
+            // load witness and shift the output by a small amount that is less than tolerance percent
+            let witness = GraphWitness::from_path(
+                format!("{}/{}/witness.json", test_dir, example_name).into(),
+            )
+            .unwrap();
+            let witness = witness.clone();
+            let outputs = witness.outputs.clone();
+
+            // get values as i128
+            let output_perturbed_safe: Vec<Vec<halo2curves::bn256::Fr>> = outputs
+                .iter()
+                .map(|sv| {
+                    sv.iter()
+                        .map(|v| {
+                            // randomly perturb by a small amount less than tolerance
+                            let perturbation = if v == &halo2curves::bn256::Fr::zero() {
+                                halo2curves::bn256::Fr::zero()
+                            } else {
+                                i128_to_felt(
+                                    (felt_to_i128(*v) as f32
+                                        * (rand::thread_rng().gen_range(-0.01..0.01) * tolerance))
+                                        as i128,
+                                )
+                            };
+
+                            *v + perturbation
+                        })
+                        .collect::<Vec<_>>()
+                })
+                .collect::<Vec<_>>();
+
+            // get values as i128
+            let output_perturbed_bad: Vec<Vec<halo2curves::bn256::Fr>> = outputs
+                .iter()
+                .map(|sv| {
+                    sv.iter()
+                        .map(|v| {
+                            // randomly perturb by a small amount less than tolerance
+                            let perturbation = if v == &halo2curves::bn256::Fr::zero() {
+                                halo2curves::bn256::Fr::from(2)
+                            } else {
+                                i128_to_felt(
+                                    (felt_to_i128(*v) as f32
+                                        * (rand::thread_rng().gen_range(0.02..0.1) * tolerance))
+                                        as i128,
+                                )
+                            };
+                            *v + perturbation
+                        })
+                        .collect::<Vec<_>>()
+                })
+                .collect::<Vec<_>>();
+
+            let good_witness = GraphWitness {
+                outputs: output_perturbed_safe,
+                ..witness.clone()
+            };
+
+            // save
+            good_witness
+                .save(format!("{}/{}/witness_ok.json", test_dir, example_name).into())
+                .unwrap();
+
+            let bad_witness = GraphWitness {
+                outputs: output_perturbed_bad,
+                ..witness.clone()
+            };
+
+            // save
+            bad_witness
+                .save(format!("{}/{}/witness_bad.json", test_dir, example_name).into())
+                .unwrap();
+
+            let status = Command::new(format!("{}/release/ezkl", *CARGO_TARGET_DIR))
+                .args([
+                    "mock",
+                    "-W",
+                    format!("{}/{}/witness.json", test_dir, example_name).as_str(),
+                    "-M",
+                    format!("{}/{}/network.compiled", test_dir, example_name).as_str(),
+                ])
+                .status()
+                .expect("failed to execute process");
+            assert!(status.success());
+
+            let status = Command::new(format!("{}/release/ezkl", *CARGO_TARGET_DIR))
+                .args([
+                    "mock",
+                    "-W",
+                    format!("{}/{}/witness_ok.json", test_dir, example_name).as_str(),
+                    "-M",
+                    format!("{}/{}/network.compiled", test_dir, example_name).as_str(),
+                ])
+                .status()
+                .expect("failed to execute process");
+            assert!(status.success());
+
+            let status = Command::new(format!("{}/release/ezkl", *CARGO_TARGET_DIR))
+                .args([
+                    "mock",
+                    "-W",
+                    format!("{}/{}/witness_bad.json", test_dir, example_name).as_str(),
+                    "-M",
+                    format!("{}/{}/network.compiled", test_dir, example_name).as_str(),
+                ])
+                .status()
+                .expect("failed to execute process");
+            assert!(!status.success());
+        } else {
+            let status = Command::new(format!("{}/release/ezkl", *CARGO_TARGET_DIR))
+                .args([
+                    "mock",
+                    "-W",
+                    format!("{}/{}/witness.json", test_dir, example_name).as_str(),
+                    "-M",
+                    format!("{}/{}/network.compiled", test_dir, example_name).as_str(),
+                ])
+                .status()
+                .expect("failed to execute process");
+            assert!(status.success());
+        }
     }
 
     #[allow(clippy::too_many_arguments)]
@@ -1312,6 +1440,7 @@ mod native_tests {
         scales_to_use: Option<Vec<u32>>,
         num_inner_columns: usize,
         div_rebasing: bool,
+        tolerance: &mut f32,
     ) {
         let mut args = vec![
             "gen-settings".to_string(),
@@ -1321,11 +1450,12 @@ mod native_tests {
                 "--settings-path={}/{}/settings.json",
                 test_dir, example_name
             ),
-            format!("--variables=batch_size={}", batch_size),
+            format!("--variables=batch_size->{}", batch_size),
             format!("--input-visibility={}", input_visibility),
             format!("--param-visibility={}", param_visibility),
             format!("--output-visibility={}", output_visibility),
             format!("--num-inner-cols={}", num_inner_columns),
+            format!("--tolerance={}", tolerance),
         ];
 
         if div_rebasing {
@@ -1368,6 +1498,24 @@ mod native_tests {
             .expect("failed to execute process");
         assert!(status.success());
 
+        let mut settings =
+            GraphSettings::load(&format!("{}/{}/settings.json", test_dir, example_name).into())
+                .unwrap();
+
+        let any_output_scales_smol = settings.model_output_scales.iter().any(|s| *s <= 0);
+
+        if any_output_scales_smol {
+            // set the tolerance to 0.0
+            settings.run_args.tolerance = Tolerance {
+                val: 0.0.into(),
+                scale: 0.0.into(),
+            };
+            settings
+                .save(&format!("{}/{}/settings.json", test_dir, example_name).into())
+                .unwrap();
+            *tolerance = 0.0;
+        }
+
         let status = Command::new(format!("{}/release/ezkl", *CARGO_TARGET_DIR))
             .args([
                 "compile-circuit",
@@ -1402,6 +1550,7 @@ mod native_tests {
     }
 
     // Mock prove (fast, but does not cover some potential issues)
+    #[allow(clippy::too_many_arguments)]
     fn accuracy_measurement(
         test_dir: &str,
         example_name: String,
@@ -1424,6 +1573,7 @@ mod native_tests {
             None,
             2,
             div_rebasing,
+            &mut 0.0,
         );
 
         println!(
@@ -1455,7 +1605,7 @@ mod native_tests {
                 format!("{}/{}/network.onnx", test_dir, example_name).as_str(),
                 "-O",
                 format!("{}/{}/render.png", test_dir, example_name).as_str(),
-                "--lookup-range=(-32768,32768)",
+                "--lookup-range=-32768->32768",
                 "-K=17",
             ])
             .status()
@@ -1683,6 +1833,7 @@ mod native_tests {
             scales_to_use,
             num_inner_columns,
             false,
+            &mut 0.0,
         );
 
         let settings_path = format!("{}/{}/settings.json", test_dir, example_name);
@@ -1745,6 +1896,30 @@ mod native_tests {
             .status()
             .expect("failed to execute process");
         assert!(status.success());
+
+        // load settings file
+        let settings =
+            std::fs::read_to_string(settings_path.clone()).expect("failed to read settings file");
+
+        let graph_settings = serde_json::from_str::<GraphSettings>(&settings)
+            .expect("failed to parse settings file");
+
+        // get_srs for the graph_settings_num_instances
+        download_srs(graph_settings.log2_total_instances());
+
+        let status = Command::new(format!("{}/release/ezkl", *CARGO_TARGET_DIR))
+            .args([
+                "verify",
+                format!("--settings-path={}", settings_path).as_str(),
+                "--proof-path",
+                &format!("{}/{}/proof.pf", test_dir, example_name),
+                "--vk-path",
+                &format!("{}/{}/key.vk", test_dir, example_name),
+                "--reduced-srs",
+            ])
+            .status()
+            .expect("failed to execute process");
+        assert!(status.success());
     }
 
     // prove-serialize-verify, the usual full path
@@ -1760,6 +1935,7 @@ mod native_tests {
             None,
             2,
             false,
+            &mut 0.0,
         );
 
         let status = Command::new(format!("{}/release/ezkl", *CARGO_TARGET_DIR))
@@ -2036,6 +2212,7 @@ mod native_tests {
             Some(vec![4]),
             1,
             false,
+            &mut 0.0,
         );
 
         let model_path = format!("{}/{}/network.compiled", test_dir, example_name);

tests/output_comparison.py

@@ -12,7 +12,7 @@ def get_ezkl_output(witness_file, settings_file):
     outputs = witness_output['outputs']
     with open(settings_file) as f:
         settings = json.load(f)
-    ezkl_outputs = [[ezkl.string_to_float(
+    ezkl_outputs = [[ezkl.felt_to_float(
         outputs[i][j], settings['model_output_scales'][i]) for j in range(len(outputs[i]))] for i in range(len(outputs))]
     return ezkl_outputs
 
@@ -91,9 +91,7 @@ def compare_outputs(zk_output, onnx_output):
                 print("------- zk_output: ", list1_i)
                 print("------- onnx_output: ", list2_i)
 
-
-
-    return np.mean(np.abs(res))
+    return res
 
 
 if __name__ == '__main__':
@@ -113,6 +111,9 @@ if __name__ == '__main__':
     onnx_output = get_onnx_output(model_file, input_file)
     # compare the outputs
     percentage_difference = compare_outputs(ezkl_output, onnx_output)
+    mean_percentage_difference = np.mean(np.abs(percentage_difference))
+    max_percentage_difference = np.max(np.abs(percentage_difference))
     # print the percentage difference
-    print("mean percent diff: ", percentage_difference)
-    assert percentage_difference < target, "Percentage difference is too high"
+    print("mean percent diff: ", mean_percentage_difference)
+    print("max percent diff: ", max_percentage_difference)
+    assert mean_percentage_difference < target, "Percentage difference is too high"

tests/py_integration_tests.rs

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

tests/python/binding_tests.py

@@ -56,9 +56,9 @@ def test_poseidon_hash():
     Test for poseidon_hash
     """
     message = [1.0, 2.0, 3.0, 4.0]
-    message = [ezkl.float_to_string(x, 7) for x in message]
+    message = [ezkl.float_to_felt(x, 7) for x in message]
     res = ezkl.poseidon_hash(message)
-    assert ezkl.string_to_felt(
+    assert ezkl.felt_to_big_endian(
         res[0]) == "0x0da7e5e5c8877242fa699f586baf770d731defd54f952d4adeb85047a0e32f45"
 
 
@@ -70,14 +70,14 @@ def test_field_serialization():
 
     input = 890
     scale = 7
-    felt = ezkl.float_to_string(input, scale)
-    roundtrip_input = ezkl.string_to_float(felt, scale)
+    felt = ezkl.float_to_felt(input, scale)
+    roundtrip_input = ezkl.felt_to_float(felt, scale)
     assert input == roundtrip_input
 
     input = -700
     scale = 7
-    felt = ezkl.float_to_string(input, scale)
-    roundtrip_input = ezkl.string_to_float(felt, scale)
+    felt = ezkl.float_to_felt(input, scale)
+    roundtrip_input = ezkl.felt_to_float(felt, scale)
     assert input == roundtrip_input
 
 
@@ -88,12 +88,12 @@ def test_buffer_to_felts():
     buffer = bytearray("a sample string!", 'utf-8')
     felts = ezkl.buffer_to_felts(buffer)
     ref_felt_1 = "0x0000000000000000000000000000000021676e6972747320656c706d61732061"
-    assert felts == [ref_felt_1]
+    assert ezkl.felt_to_big_endian(felts[0]) == ref_felt_1
 
     buffer = bytearray("a sample string!"+"high", 'utf-8')
     felts = ezkl.buffer_to_felts(buffer)
     ref_felt_2 = "0x0000000000000000000000000000000000000000000000000000000068676968"
-    assert felts == [ref_felt_1, ref_felt_2]
+    assert [ezkl.felt_to_big_endian(felts[0]),  ezkl.felt_to_big_endian(felts[1])] == [ref_felt_1, ref_felt_2]
 
 
 def test_gen_srs():
@@ -113,13 +113,13 @@ def test_calibrate_over_user_range():
     data_path = os.path.join(
         examples_path,
         'onnx',
-        '1l_average',
+        '1l_relu',
         'input.json'
     )
     model_path = os.path.join(
         examples_path,
         'onnx',
-        '1l_average',
+        '1l_relu',
         'network.onnx'
     )
     output_path = os.path.join(
@@ -147,13 +147,13 @@ def test_calibrate():
     data_path = os.path.join(
         examples_path,
         'onnx',
-        '1l_average',
+        '1l_relu',
         'input.json'
     )
     model_path = os.path.join(
         examples_path,
         'onnx',
-        '1l_average',
+        '1l_relu',
         'network.onnx'
     )
     output_path = os.path.join(
@@ -183,7 +183,7 @@ def test_model_compile():
     model_path = os.path.join(
         examples_path,
         'onnx',
-        '1l_average',
+        '1l_relu',
         'network.onnx'
     )
     compiled_model_path = os.path.join(
@@ -205,7 +205,7 @@ def test_forward():
     data_path = os.path.join(
         examples_path,
         'onnx',
-        '1l_average',
+        '1l_relu',
         'input.json'
     )
     model_path = os.path.join(
@@ -392,9 +392,7 @@ def test_prove_evm():
     assert res['transcript_type'] == 'EVM'
     assert os.path.isfile(proof_path)
 
-    res = ezkl.print_proof_hex(proof_path)
-    # to figure out a better way of testing print_proof_hex
-    assert type(res) == str
+
 
 
 def test_create_evm_verifier():

tests/wasm.rs

@@ -8,9 +8,9 @@ mod wasm32 {
     use ezkl::graph::GraphWitness;
     use ezkl::pfsys;
     use ezkl::wasm::{
-        bufferToVecOfstring, compiledCircuitValidation, encodeVerifierCalldata, genPk, genVk,
-        genWitness, inputValidation, pkValidation, poseidonHash, printProofHex, proofValidation,
-        prove, settingsValidation, srsValidation, stringToFelt, stringToFloat, stringToInt,
+        bufferToVecOfFelt, compiledCircuitValidation, encodeVerifierCalldata, feltToBigEndian,
+        feltToFloat, feltToInt, genPk, genVk, genWitness, inputValidation, pkValidation,
+        poseidonHash, proofValidation, prove, settingsValidation, srsValidation,
         u8_array_to_u128_le, verify, vkValidation, witnessValidation,
     };
     use halo2_solidity_verifier::encode_calldata;
@@ -76,21 +76,21 @@ mod wasm32 {
         for i in 0..32 {
             let field_element = Fr::from(i);
             let serialized = serde_json::to_vec(&field_element).unwrap();
+
             let clamped = wasm_bindgen::Clamped(serialized);
             let scale = 2;
-            let floating_point = stringToFloat(clamped.clone(), scale)
+            let floating_point = feltToFloat(clamped.clone(), scale)
                 .map_err(|_| "failed")
                 .unwrap();
             assert_eq!(floating_point, (i as f64) / 4.0);
 
-            let integer: i128 = serde_json::from_slice(
-                &stringToInt(clamped.clone()).map_err(|_| "failed").unwrap(),
-            )
-            .unwrap();
+            let integer: i128 =
+                serde_json::from_slice(&feltToInt(clamped.clone()).map_err(|_| "failed").unwrap())
+                    .unwrap();
             assert_eq!(integer, i as i128);
 
             let hex_string = format!("{:?}", field_element);
-            let returned_string = stringToFelt(clamped).map_err(|_| "failed").unwrap();
+            let returned_string: String = feltToBigEndian(clamped).map_err(|_| "failed").unwrap();
             assert_eq!(hex_string, returned_string);
         }
     }
@@ -101,7 +101,7 @@ mod wasm32 {
         let mut buffer = string_high.clone().into_bytes();
         let clamped = wasm_bindgen::Clamped(buffer.clone());
 
-        let field_elements_ser = bufferToVecOfstring(clamped).map_err(|_| "failed").unwrap();
+        let field_elements_ser = bufferToVecOfFelt(clamped).map_err(|_| "failed").unwrap();
 
         let field_elements: Vec<Fr> = serde_json::from_slice(&field_elements_ser[..]).unwrap();
 
@@ -118,7 +118,7 @@ mod wasm32 {
         let buffer = string_sample.clone().into_bytes();
         let clamped = wasm_bindgen::Clamped(buffer.clone());
 
-        let field_elements_ser = bufferToVecOfstring(clamped).map_err(|_| "failed").unwrap();
+        let field_elements_ser = bufferToVecOfFelt(clamped).map_err(|_| "failed").unwrap();
 
         let field_elements: Vec<Fr> = serde_json::from_slice(&field_elements_ser[..]).unwrap();
 
@@ -133,7 +133,7 @@ mod wasm32 {
         let buffer = string_concat.into_bytes();
         let clamped = wasm_bindgen::Clamped(buffer.clone());
 
-        let field_elements_ser = bufferToVecOfstring(clamped).map_err(|_| "failed").unwrap();
+        let field_elements_ser = bufferToVecOfFelt(clamped).map_err(|_| "failed").unwrap();
 
         let field_elements: Vec<Fr> = serde_json::from_slice(&field_elements_ser[..]).unwrap();
 
@@ -258,15 +258,6 @@ mod wasm32 {
         assert!(value);
     }
 
-    #[wasm_bindgen_test]
-    async fn print_proof_hex_test() {
-        let proof = printProofHex(wasm_bindgen::Clamped(PROOF.to_vec()))
-            .map_err(|_| "failed")
-            .unwrap();
-
-        assert!(proof.len() > 0);
-    }
-
     #[wasm_bindgen_test]
     async fn verify_validations() {
         // Run witness validation on network (should fail)

tests/wasm/witness.json

@@ -1 +1 @@
-{"inputs":[["0200000000000000000000000000000000000000000000000000000000000000","0100000000000000000000000000000000000000000000000000000000000000","0100000000000000000000000000000000000000000000000000000000000000"]],"pretty_elements":{"rescaled_inputs":[["2","1","1"]],"inputs":[["0x0000000000000000000000000000000000000000000000000000000000000002","0x0000000000000000000000000000000000000000000000000000000000000001","0x0000000000000000000000000000000000000000000000000000000000000001"]],"processed_inputs":[],"processed_params":[],"processed_outputs":[],"rescaled_outputs":[["0","0","0","0"]],"outputs":[["0x0000000000000000000000000000000000000000000000000000000000000000","0x0000000000000000000000000000000000000000000000000000000000000000","0x0000000000000000000000000000000000000000000000000000000000000000","0x0000000000000000000000000000000000000000000000000000000000000000"]]},"outputs":[["0000000000000000000000000000000000000000000000000000000000000000","0000000000000000000000000000000000000000000000000000000000000000","0000000000000000000000000000000000000000000000000000000000000000","0000000000000000000000000000000000000000000000000000000000000000"]],"processed_inputs":null,"processed_params":null,"processed_outputs":null,"max_lookup_inputs":0,"min_lookup_inputs":-1}
+{"inputs":[["0200000000000000000000000000000000000000000000000000000000000000","0100000000000000000000000000000000000000000000000000000000000000","0100000000000000000000000000000000000000000000000000000000000000"]],"pretty_elements":{"rescaled_inputs":[["2","1","1"]],"inputs":[["0x0000000000000000000000000000000000000000000000000000000000000002","0x0000000000000000000000000000000000000000000000000000000000000001","0x0000000000000000000000000000000000000000000000000000000000000001"]],"processed_inputs":[],"processed_params":[],"processed_outputs":[],"rescaled_outputs":[["0","0","0","0"]],"outputs":[["0x0000000000000000000000000000000000000000000000000000000000000000","0x0000000000000000000000000000000000000000000000000000000000000000","0x0000000000000000000000000000000000000000000000000000000000000000","0x0000000000000000000000000000000000000000000000000000000000000000"]]},"outputs":[["0000000000000000000000000000000000000000000000000000000000000000","0000000000000000000000000000000000000000000000000000000000000000","0000000000000000000000000000000000000000000000000000000000000000","0000000000000000000000000000000000000000000000000000000000000000"]],"processed_inputs":null,"processed_params":null,"processed_outputs":null,"max_lookup_inputs":0,"min_lookup_inputs":-1,"max_range_size":0}