Shumai

Shumai is a tensor library that enables JavaScript developers to build and train machine learning models with production-grade performance. It bridges the gap between web development and numerical computing by providing differentiable tensors, automatic gradient calculation, and efficient matrix operations.

Highlights

• Fast execution through FFI bindings with Bun runtime
• GPU acceleration support for large-scale computations
• Automatic differentiation for backpropagation and gradient tracking
• Network connectivity for distributed ML workflows
• Standard tensor operations and neural network primitives
• TypeScript-first design with strong type safety

The library targets JavaScript developers who need serious ML capabilities without switching languages. It's particularly useful for integrating models into web applications, building data processing pipelines in Node.js, and experimenting with neural networks in a language most frontend and full-stack engineers already know. The project remains experimental but actively maintained, making it suitable for development and research use cases.