Skip to content

zant_sh_readme.md

Latest commit

 

History

History
104 lines (76 loc) · 2.87 KB

zant_sh_readme.md

File metadata and controls

104 lines (76 loc) · 2.87 KB
Raw

Zant - ONNX Python Tools Wrapper

A shell script wrapper for various ONNX Python tools that provides a unified interface with proper flag handling and help documentation.

Installation

  1. Make the script executable:
chmod +x zant
  1. Optionally, add it to your PATH or create a symlink:
# Option 1: Copy to a directory in your PATH
sudo cp zant /usr/local/bin/

# Option 2: Create a symlink
ln -s $(pwd)/zant /usr/local/bin/zant

Usage

./zant <script> [flags]

Available Scripts

  • create - runs all the below, to use just once all the below are tested
  • onnx_gen - Generate fuzzed ONNX models and save execution data in JSON
  • onnx_extract - For a N nodes model it creates N onnx models, one for each node with respective tests.
  • user_tests_gen - Run ONNX model multiple times with random inputs and save execution data
  • shape_thief - Upgrade your model with all intermediate tensor's shapes
  • input_setter - Set input shape and infer shapes of the ONNX model

Getting Help

# General help
./zant --help

# Script-specific help
./zant <script> --help

Examples

See zant_CLI for more details!

Script Details

onnx_gen

Location: tests/CodeGen/Python-ONNX/onnx_gen.py

Flags:

  • --iterations - Number of models to generate for each operation (default: 1)
  • --seed - Seed for random generation (for reproducibility)
  • --output-dir - Directory to save generated models (default: datasets/oneOpModels)
  • --metadata-file - File to save metadata and execution data (default: datasets/oneOpModels/results.json)
  • --op - Name of the operation you want to generate and test (default: all)

user_tests_gen

Location: tests/CodeGen/user_tests_gen.py

Flags:

  • --model - Your ONNX model (required)
  • --iterations - Number of randomized inference runs (default: 1)

shape_thief

Location: src/onnx/shape_thief.py

input_setter

Location: src/onnx/input_setter.py

Flags:

  • --model - Name of your model (required)
  • --shape - Input shape as comma-separated values, e.g., 1,3,34,34 (required)

Requirements

  • Python 3 (or Python 2 as fallback)
  • The respective Python scripts must exist in their specified locations
  • Required Python dependencies for each script (ONNX, numpy, etc.)

Error Handling

The script includes comprehensive error handling:

  • Validates script names
  • Checks for Python script file existence
  • Verifies Python interpreter availability
  • Provides helpful error messages with usage instructions

Directory Structure

The script expects the following directory structure:

project/
├── zant                                    # This wrapper script
├── tests/CodeGen/Python-ONNX/onnx_gen.py
├── tests/CodeGen/user_tests_gen.py
└── src/onnx/
    ├── shape_thief.py
    └── input_setter.py