FHE Benchmarking Suite - ML Inference

This repository contains the harness for the ML-inference workload of the FHE benchmarking suite of HomomorphicEncryption.org. The harness currently supports mnist model benchmarking as specified in harness/mnist directory. The main branch contains a reference implementation of this workload, under the submission subdirectory. The harness also supports an optional remote backend execution mode under the submission_remote subdirectory, where the homomorphic evaluation is executed on a remote backend.

Submitters need to clone this repository, replace the content of the submission or submission_remote subdirectory by their own implementation. They also may need to changes or replace the script scripts/build_task.sh to account for dependencies and build environment for their submission. Submitters are expected to document any changes made to the model architecture harness/mnist/mnist.py in the submission/README.md file.

Execution Modes

The ML Inference benchmark supports two execution modes:

Local Execution (Default)

All steps are executed on a single machine:

• Cryptographic context setup and model preprocessing
• Key generation
• Input preprocessing and encryption
• Homomorphic inference
• Decryption and postprocessing

This corresponds to the reference submission in submission/.

Remote Backend Execution (Optional)

Some FHE deployments separate client-side and server-side responsibilities.
In this mode:

• Client-side (local):

  • Key generation
  • Input preprocessing and encryption
  • Decryption and postprocessing

• Server-side (remote):

  • Cryptographic context setup and model preprocessing
  • Homomorphic inference

This execution mode is enabled by passing the --remote flag to the harness.

Running the ML-inference workload

Dependencies

• Python 3.12+
• The build environment for local execution depends on OpenFHE being installed as specificied in scripts/get_openfhe.sh and submission/CMakeLists.txt. See https://github.com/openfheorg/openfhe-development#installation.
• The build environment for remote-backend execution depends on lattica-query being installed as specified in submission_remote/requirements.txt. See https://platformdocs.lattica.ai/how-to-guides/client-installation/how-to-install-query-client. Should be installed on a linux_x86_64 machine.

Execution

To run the workload, clone and install dependencies:

git clone https://github.com/fhe-benchmarking/ml-inference.git
cd ml-inference

python -m venv bmenv
source ./bmenv/bin/activate
pip install -r requirements.txt

python3 harness/run_submission.py -h  # Information about command-line options

The harness script harness/run_submission.py will attempt to build the submission itself, if it is not already built. If already built, it will use the same project without re-building it (unless the code has changed). An example run is provided below.

$ python3 harness/run_submission.py -h
usage: run_submission.py [-h] [--num_runs NUM_RUNS] [--seed SEED] [--clrtxt CLRTXT] [--remote] {0,1,2,3}

Run ML Inference FHE benchmark.

positional arguments:
  {0,1,2,3}            Instance size (0-single/1-small/2-medium/3-large)

options:
  -h, --help           show this help message and exit
  --num_runs NUM_RUNS  Number of times to run steps 4-9 (default: 1)
  --seed SEED          Random seed for dataset and query generation
  --clrtxt CLRTXT      Specify with 1 if to rerun the cleartext computation
  --remote             Specify if to run in remote-backend mode

The single instance runs the inference for a single input and verifies the correctness of the obtained label compared to the ground-truth label.

$ python3 ./harness/run_submission.py 0 --seed 3 --num_runs 2
 

[harness] Running submission for single inference
[get-openfhe] Found OpenFHE at .../ml-inference/third_party/openfhe (use --force to rebuild).
-- FOUND PACKAGE OpenFHE
-- OpenFHE Version: 1.3.1
-- OpenFHE installed as shared libraries: ON
-- OpenFHE include files location: .../ml-inference/third_party/openfhe/include/openfhe
-- OpenFHE lib files location: .../ml-inference/third_party/openfhe/lib
-- OpenFHE Native Backend size: 64
-- Configuring done (0.0s)
-- Generating done (0.0s)
-- Build files have been written to: .../ml-inference/submission/build
[ 12%] Built target client_preprocess_input
[ 25%] Built target client_postprocess
[ 37%] Built target server_preprocess_model
[ 62%] Built target client_key_generation
[ 62%] Built target mlp_encryption_utils
[ 75%] Built target client_encode_encrypt_input
[100%] Built target client_decrypt_decode
[100%] Built target server_encrypted_compute
22:50:49 [harness] 1: Harness: MNIST Test dataset generation completed (elapsed: 7.5552s)
22:50:51 [harness] 2: Client: Key Generation completed (elapsed: 2.2688s)
         [harness] Client: Public and evaluation keys size: 1.4G
22:50:51 [harness] 3: Server: (Encrypted) model preprocessing completed (elapsed: 0.1603s)

         [harness] Run 1 of 2
100.0%
100.0%
100.0%
100.0%
22:51:04 [harness] 4: Harness: Input generation for MNIST completed (elapsed: 13.1305s)
22:51:04 [harness] 5: Client: Input preprocessing completed (elapsed: 0.0489s)
22:51:04 [harness] 6: Client: Input encryption completed (elapsed: 0.0481s)
         [harness] Client: Encrypted input size: 358.8K
         [server] Loading keys
         [server] Run encrypted MNIST inference
         [server] Execution time for ciphertext 0 : 11 seconds
22:51:18 [harness] 7: Server: Encrypted ML Inference computation completed (elapsed: 13.3027s)
         [harness] Client: Encrypted results size: 69.6K
22:51:18 [harness] 8: Client: Result decryption completed (elapsed: 0.1729s)
22:51:18 [harness] 9: Client: Result postprocessing completed (elapsed: 0.0921s)
[harness] PASS  (expected=7, got=7)
[total latency] 36.7796s

         [harness] Run 2 of 2
22:51:23 [harness] 4: Harness: Input generation for MNIST completed (elapsed: 5.2028s)
22:51:23 [harness] 5: Client: Input preprocessing completed (elapsed: 0.0986s)
22:51:23 [harness] 6: Client: Input encryption completed (elapsed: 0.0998s)
         [harness] Client: Encrypted input size: 358.8K
         [server] Loading keys
         [server] Run encrypted MNIST inference
         [server] Execution time for ciphertext 0 : 12 seconds
22:51:37 [harness] 7: Server: Encrypted ML Inference computation completed (elapsed: 13.8138s)
         [harness] Client: Encrypted results size: 69.6K
22:51:37 [harness] 8: Client: Result decryption completed (elapsed: 0.1219s)
22:51:37 [harness] 9: Client: Result postprocessing completed (elapsed: 0.0827s)
[harness] PASS  (expected=7, got=7)
[total latency] 29.4041s

All steps completed for the single inference!

The batch inference cases run the inference for a batch of inputs of varying sizes. The accuracy (with respect to the ground truth labels) is compared between the decrypted results and the results obtained using the harness model.

$python3 ./harness/run_submission.py 1 --seed 3 --num_runs 2

[harness] Running submission for small inference
[harness] Running submission for single inference
[get-openfhe] Found OpenFHE at .../ml-inference/third_party/openfhe (use --force to rebuild).
-- FOUND PACKAGE OpenFHE
-- OpenFHE Version: 1.3.1
-- OpenFHE installed as shared libraries: ON
-- OpenFHE include files location: .../ml-inference/third_party/openfhe/include/openfhe
-- OpenFHE lib files location: .../ml-inference/third_party/openfhe/lib
-- OpenFHE Native Backend size: 64
-- Configuring done (0.0s)
-- Generating done (0.0s)
-- Build files have been written to: .../ml-inference/submission/build
[ 12%] Built target client_preprocess_input
[ 25%] Built target client_postprocess
[ 37%] Built target server_preprocess_model
[ 62%] Built target client_key_generation
[ 62%] Built target mlp_encryption_utils
[ 75%] Built target client_encode_encrypt_input
[100%] Built target client_decrypt_decode
[100%] Built target server_encrypted_compute
22:44:03 [harness] 1: Harness: MNIST Test dataset generation completed (elapsed: 7.5536s)
22:44:05 [harness] 2: Client: Key Generation completed (elapsed: 2.1305s)
         [harness] Client: Public and evaluation keys size: 1.4G
22:44:05 [harness] 3: Server: (Encrypted) model preprocessing completed (elapsed: 0.1265s)

         [harness] Run 1 of 2
22:44:08 [harness] 4: Harness: Input generation for MNIST completed (elapsed: 3.2961s)
22:44:08 [harness] 5: Client: Input preprocessing completed (elapsed: 0.0879s)
22:44:09 [harness] 6: Client: Input encryption completed (elapsed: 0.1254s)
         [harness] Client: Encrypted input size: 5.2M
         [server] Loading keys
         [server] Run encrypted MNIST inference
         [server] Execution time for ciphertext 0 : 9 seconds
         [server] Execution time for ciphertext 1 : 7 seconds
         [server] Execution time for ciphertext 2 : 7 seconds
         [server] Execution time for ciphertext 3 : 8 seconds
         [server] Execution time for ciphertext 4 : 8 seconds
         [server] Execution time for ciphertext 5 : 8 seconds
         [server] Execution time for ciphertext 6 : 8 seconds
         [server] Execution time for ciphertext 7 : 8 seconds
         [server] Execution time for ciphertext 8 : 8 seconds
         [server] Execution time for ciphertext 9 : 8 seconds
         [server] Execution time for ciphertext 10 : 8 seconds
         [server] Execution time for ciphertext 11 : 8 seconds
         [server] Execution time for ciphertext 12 : 8 seconds
         [server] Execution time for ciphertext 13 : 8 seconds
         [server] Execution time for ciphertext 14 : 9 seconds
22:46:17 [harness] 7: Server: Encrypted ML Inference computation completed (elapsed: 128.6067s)
         [harness] Client: Encrypted results size: 988.6K
22:46:17 [harness] 8: Client: Result decryption completed (elapsed: 0.2126s)
22:46:17 [harness] 9: Client: Result postprocessing completed (elapsed: 0.1055s)
22:46:23 [harness] 10.1: Harness: Run inference for harness plaintext model completed (elapsed: 5.1714s)
         [harness] Wrote harness model predictions to:  .../ml-inference/io/small/harness_model_predictions.txt
[harness] Encrypted Model Accuracy: 0.9333 (14/15 correct)
[harness] Harness Model Accuracy: 0.9333 (14/15 correct)
22:46:23 [harness] 10.2: Harness: Run quality check on encrypted inference completed (elapsed: 0.0008s)
[total latency] 147.4171s

         [harness] Run 2 of 2
22:46:26 [harness] 4: Harness: Input generation for MNIST completed (elapsed: 3.51s)
22:46:26 [harness] 5: Client: Input preprocessing completed (elapsed: 0.1004s)
22:46:26 [harness] 6: Client: Input encryption completed (elapsed: 0.1497s)
         [harness] Client: Encrypted input size: 5.2M
         [server] Loading keys
         [server] Run encrypted MNIST inference
         [server] Execution time for ciphertext 0 : 11 seconds
         [server] Execution time for ciphertext 1 : 8 seconds
         [server] Execution time for ciphertext 2 : 8 seconds
         [server] Execution time for ciphertext 3 : 8 seconds
         [server] Execution time for ciphertext 4 : 8 seconds
         [server] Execution time for ciphertext 5 : 8 seconds
         [server] Execution time for ciphertext 6 : 8 seconds
         [server] Execution time for ciphertext 7 : 8 seconds
         [server] Execution time for ciphertext 8 : 8 seconds
         [server] Execution time for ciphertext 9 : 8 seconds
         [server] Execution time for ciphertext 10 : 8 seconds
         [server] Execution time for ciphertext 11 : 8 seconds
         [server] Execution time for ciphertext 12 : 8 seconds
         [server] Execution time for ciphertext 13 : 8 seconds
         [server] Execution time for ciphertext 14 : 8 seconds
22:48:38 [harness] 7: Server: Encrypted ML Inference computation completed (elapsed: 131.3166s)
         [harness] Client: Encrypted results size: 988.6K
22:48:38 [harness] 8: Client: Result decryption completed (elapsed: 0.2358s)
22:48:38 [harness] 9: Client: Result postprocessing completed (elapsed: 0.085s)
22:48:43 [harness] 10.1: Harness: Run inference for harness plaintext model completed (elapsed: 4.9384s)
         [harness] Wrote harness model predictions to:  .../ml-inference/io/small/harness_model_predictions.txt
[harness] Encrypted Model Accuracy: 0.9333 (14/15 correct)
[harness] Harness Model Accuracy: 0.9333 (14/15 correct)
22:48:43 [harness] 10.2: Harness: Run quality check on encrypted inference completed (elapsed: 0.0007s)
[total latency] 150.1474s

All steps completed for the small inference!

After finishing the run, deactivate the virtual environment.

deactivate

Directory structure

The directory structure of this reposiroty is as follows:

├─ README.md     # This file
├─ LICENSE.md    # Harness software license (Apache v2)
├─ harness/      # Scripts to drive the workload implementation
|   ├─ run_submission.py
|   ├─ verify_result.py
|   ├─ calculate_quality.py
|   └─ [...]
├─ datasets/     # The harness scripts create and populate this directory
├─ docs/         # Optional: additional documentation
├─ io/           # This directory is used for client<->server communication
├─ measurements/ # Holds logs with performance numbers
├─ scripts/      # Helper scripts for dependencies and build system
└─ submission/   # This is where the workload implementation lives
    ├─ README.md   # Submission documentation (mandatory)
    ├─ LICENSE.md  # Optional software license (if different from Apache v2)
    └─ [...]
└─ submission_remote/  # This is where the remote-backend workload implementation lives
    └─ [...]

Submitters must overwrite the contents of the scripts and submissions subdirectories.

Description of stages

A submitter can edit any of the client_* / server_* sources in /submission. Moreover, for the particular parameters related to a workload, the submitter can modify the params files. If the current description of the files are inaccurate, the stage names in run_submission can be also modified.

The current stages are the following, targeted to a client-server scenario. The order in which they are happening in run_submission assumes an initialization step which is database-dependent and run only once, and potentially multiple runs for multiple queries. Each file can take as argument the test case size.

Stage executables	Description
server_get_params	(Optional) Get cryptographic context from a remote server.
client_key_generation	Generate all key material and cryptographic context at the client.
server_upload_ek	(Optional) Upload evaluation key to a remote backend.
client_preprocess_dataset	(Optional) Any in the clear computations the client wants to apply over the dataset/model.
client_preprocess_input	(Optional) Any in the clear computations the client wants to apply over the input.
client_encode_encrypt_query	Plaintext encoding and encryption of the input at the client.
server_preprocess_model	(Optional) Any in the clear or encrypted computations the server wants to apply over the model.
server_encrypted_compute	The computation the server applies to achieve the workload solution over encrypted data.
client_decrypt_decode	Decryption and plaintext decoding of the result at the client.
client_postprocess	Any in the clear computation that the client wants to apply on the decrypted result.

The outer python script measures the runtime of each stage. The current stage separation structure requires reading and writing to files more times than minimally necessary. For a more granular runtime measuring, which would account for the extra overhead described above, we encourage submitters to separate and print in a log the individual times for reads/writes and computations inside each stage.