MoonTensorLite

Project identifier: moonbit/moontensorlite Author: Qlc67800

MoonTensorLite is a lightweight MoonBit tensor and automatic differentiation teaching framework. It is designed to stay small, readable, and testable while still supporting the core building blocks needed for linear regression, MLPs, and softmax classification.

Repository

• GitLink: https://gitlink.org.cn/Qlc67800/moontensorlite
• GitHub: https://github.com/qlcooo677/moontensorlite

Source note

This repository is authored from scratch for the MoonBit open-source contest. The code will continue to be expanded in public, with clear commit history, tests, and documentation.

What this project aims to do

• Provide a compact tensor core with shape-aware operations
• Implement reverse-mode autodiff with clear computation graph boundaries
• Offer a tiny neural network layer for teaching and competition demos
• Keep the codebase easy to inspect, benchmark, and extend

Current status

This repository is being prepared for the 2026 MoonBit open source competition. The initial scaffold is in place and the implementation will grow in small, reviewable steps.

Build and test

moon check
moon test
moon run cmd/main

Planned scope

• Tensor creation and basic math
• Broadcasting and matrix multiplication
• Backpropagation for scalar and tensor expressions
• Minimal training examples for regression and classification