MLVisualizer

This project provides a command-line tool to parse machine learning models, benchmark them on CPU/GPU with datasets, and visualize their architecture using Netron.

Features

• Model Parsing
  Extract model information using MLParser.

• Benchmarking
  Run performance tests with custom datasets on CPU and GPU via BenchMark.

• Visualization
  Launch a local Netron instance to inspect model architecture.

• CLI Interface
  Simple interface with options for metrics, dataset benchmarking, and visualization.

Installation

1. Clone this repository:

   • Make sure to install Poetry: https://python-poetry.org/

   git clone https://github.com/marcellobeltrami/mlInspect.git
   cd mlInspect
   poetry install
   poetry run python src/mlinspect/main.py -h # prints help message

2. Run the project:

   poetry run python main.py --input_model <path_to_model> [options]

   Options

   --input_model	Path to the ML model file	Required
   --metrics	Show model information parsed by MLParser	False
   --dataset_encoded_cpu	Path to encoded dataset for CPU benchmarking	False
   --dataset_encoded_gpu	Path to encoded dataset for GPU benchmarking	False
   --visualize	Launch Netron to visualize the model

   Limitations

   • Model supported are .keras, .h5, .pth and .ph
   • encoded datasets must be encoded .npy files saved from numpy. See tests/ for example input files.

Structure

├── mvp/
│   ├── ml_parser.py      # Model parsing logic
│   ├── cli.py            # CLI handler (UX)
├── visual/
│   └── view.py           # Visualization wrapper
├── benchmark/
│   └── model_benc.py     # Benchmarking module
├── main.py               # Entry point
├── requirements.txt
└── README.md

Requirements

python = ">=3.12,<3.14"
torch = "2.8.0"
tensorflow = "2.20.0"
tabulate = "0.9.0"
pandas = "2.3.2"
netron = "8.6.5"
numpy = "2.3.3"