Introduction

torch is a lightweight machine learning library for Source Academy, inspired by PyTorch. It is written in TypeScript/JavaScript and exposed to Python through a Pyodide bridge.

Why torch?

Source Academy runs in the browser. Real PyTorch cannot run in WebAssembly. torch fills that gap — it provides a familiar, PyTorch-like API for students learning machine learning concepts without requiring a local Python installation.

Architecture

┌─────────────────────────────────┐
│        Python (Pyodide)         │
│   bridge.py  ←  your code       │
└────────────┬────────────────────┘
             │  JS ↔ Python FFI
┌────────────▼────────────────────┐
│    TypeScript / JavaScript      │
│         veehz/torch             │
└─────────────────────────────────┘

The TypeScript core handles all computation. bridge.py wraps the JS objects in a Pythonic interface so you can write code that looks like standard PyTorch.