facebookresearch/shumai

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1172 stars · 25 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Shumai is a fast, differentiable tensor library for TypeScript and JavaScript built on Bun and Flashlight (ArrayFire backend). It provides N-dimensional math with automatic differentiation, GPU acceleration, and network-connected tensors for machine learning research in JS.

Use cases

• run tensor math and autograd in TypeScript
• train neural networks from JavaScript with Bun
• do GPU-accelerated N-dimensional array computations in JS
• compute gradients with automatic differentiation in TypeScript
• share tensors over the network between processes
• prototype ML models without leaving the JavaScript ecosystem

When to choose

• you want differentiable tensor ops in TypeScript/JavaScript with Bun
• you need GPU-backed tensor math via ArrayFire on macOS or Linux
• you're a JS developer experimenting with ML or gradient-based optimization

When to avoid

• you need a mature, production-grade deep learning framework
• you require Windows support or broad platform coverage
• your project depends on the PyTorch/TensorFlow ecosystem or extensive model zoos

Facets

library · maturity experimental

machine-learning deep-learning llm-training math networking machine-learning deep-learning developer-tools cross-platform bun tensor-library autograd automatic-differentiation gpu-accelerated arrayfire flashlight bun macos linux typescript nodejs

Member repositories

For agents

markdown · JSON · MCP: product_card(name="facebookresearch/shumai")