sonusai-torchl 1.3.0

PyTorch Lightning model tools for SonusAI

SonusAI Torchl

PyTorch Lightning model tools for SonusAI.

SonusAI Torchl is a companion project to SonusAI, providing tools and utilities for training, predicting, and exporting deep neural network models using PyTorch Lightning.

Key Features

• Lightning Training: Streamlined training workflows using PyTorch Lightning.
• Model Export: Tools for converting trained models to ONNX and PyTorch AOTInductor (.so) formats.
• Data Loading: Specialized data generators and datasets compatible with SonusAI mixture databases.
• Inference: Prediction tools for running inference on trained Lightning models.

Getting Started

Refer to the Development Guide for installation and setup instructions.

Prerequisites

• Python 3.13+ (Python 3.13 and 3.14 supported)
• sonusai
• uv

Quick Install

If you have uv installed, you can sync the environment:

uv sync

Documentation

• CLI Reference: Overview of available sonusai_torchl commands.
• Development Guide: Setup, build, and test instructions.
• Data Formats & Dimensions: Technical specifications for data processing.
• Architecture: Details on the project structure and integration.

Command Overview

   torchl2aoti                  Convert a trained PyTorch Lightning model to AOTInductor
   torchl_onnx                  Convert a trained PyTorch Lightning model to ONNX
   torchl_predict               Run PyTorch Lightning predict on a trained model
   torchl_train                 Train a model using PyTorch Lightning

Use python -m sonusai_torchl.<command> --help for detailed information on any command.