ExecuTorch

Docs

0. Main Page
1. Executorch 오픈소스 분석
2. 모델 실행 검증 도구
3. 실행 로깅 유틸리티

---

Project Structure

├── README.md
├── benchmark
│   └── run_benchmark.py # logging benchmark
├── model
│   ├── export_resnet18.py # model export
│   └── model_outputs # executorch model dir
│       └── resnet18.pte
├── pyproject.toml # setting for run_bench cli
├── requirements.txt
└── tests # pytests
    ├── test_benchmark_logs.py # test run_benchmark
    └── test_resnet18_equivalence.py # compare pytorch & executorch model

---

Quick Start

0. Recommended Env

# craete venv with Python 3.10
conda create -n python310 python=3.10

# activate venv
conda activate python310
# verify python version 
python --version

1. Git Clone

git clone https://github.com/SeoyeonPark1223/EXECUTORCH.git
cd EXECUTORCH

2. Installation

requirement.txt

pip install -r requirements.txt

pyproject.toml for run_bench cli

pip install .

3. Test

Run Pytest

pytest -s tests/

Run CLI Benchmark (example)

run_bench --help
run_bench --model {model_name} --repeat {repeat_num}

---

Test Examples

Pytest 1:

pytest -s tests/test_resnet18_equivalence()

Example output:

model_name: resnet18
mean_absolute_difference: 0.00000121
pytorch_latency: avg 14.77 ms, max 30.64 ms
executorch_latency: avg 11.44 ms, max 14.74 ms
====== 1 passed, 9 warnings in 10.07s ======

Pytest 2:

pytest -s tests/test_benchmark_logs.py

Example output:

{
"model_name": "resnet18.pte",
"latency_ms_avg": 13.76,
"repeat": 5
}
====== 1 passed, 1 warning in 2.69s ======

CLI:

run_bench --model resnet18.pte --repeat 5

Example output:

{
"model_name": "resnet18.pte",
"latency_ms_avg": 12.98,
"repeat": 5
}

---

Reference

• Executorch GitHub
• PyTorch Docs
• Introducing ExecuTorch from PyTorch Edge (YouTube)
• Export API Reference
• Runtime Python API Reference