Shumai

Active

Tagline:Fast differentiable tensor library for TypeScript and JavaScript with Bun + Flashlight

Updated:-

Social Media:

AI Developer Tools Research & Data Analysis Open Source Free For Developers Linux Compatible Mac Compatible Self-Hosted

Open Website

Visit Website

What is Shumai?

Shumai is a fast, network-connected, differentiable tensor library for TypeScript and JavaScript, built with Bun and Flashlight for software engineers and researchers. Create tensors, run matrix math, compute gradients with autograd, and serve models over the network. Linux installs default to GPU computation with CUDA; macOS defaults to CPU via ArrayFire. MIT licensed open-source project from Meta AI Research.

How to use Shumai?

Install Bun and ArrayFire (brew install arrayfire on macOS, or apt packages on Ubuntu). Run bun install @shumai/shumai. Import with import * as sm from '@shumai/shumai', create tensors with sm.randn() or sm.tensor(), and call .backward() for gradients. See docs at facebookresearch.github.io/shumai for networked tensor examples.

Shumai Core Features

Differentiable tensors with autograd

Set requires_grad on tensors and call backward() to populate gradients automatically. Supports detach() to copy tensors without gradient tracking.

Network-connected tensors

Serve models with sm.io.serve_model and call remote tensors from clients via sm.io.remote_model for distributed ML over the network.

Bun + Flashlight backend

Built on Bun runtime and Flashlight with ArrayFire. Linux defaults to CUDA GPU; macOS uses ArrayFire CPU backend. Standard ops include matmul, randn, identity, and loss functions.

JavaScript native arrays

Convert between Float32Array and tensors with toFloat32Array() and sm.tensor(). Scalar tensors convert to JavaScript numbers via toFloat32().

Shumai Use Cases

Prototype ML experiments in TypeScript without leaving the JS ecosystem

Build networked ML services with remote tensor calls and server-side autograd

Research differentiable programming with gradient-based optimization in Bun

Shumai Frequently Asked Questions

Which platforms are supported?

macOS and Linux are officially supported. Windows users can use Docker + WSL2 + Linux, including CUDA support.

How do I install Shumai?

Install Bun and ArrayFire, then run bun install @shumai/shumai. macOS users should install ArrayFire CPU via Homebrew; Linux users should install the CUDA backend via apt on Ubuntu.

Is Shumai production-ready?

This is experimental software. The README warns it is experimental. File issues on GitHub if you encounter problems with install or performance.

Shumai Product Background & Official Links

Shumai Support Email & Customer service contact & Refund contact etc.

More Contact, visit the about us page(https://facebookresearch.github.io/shumai/)

Shumai Company Info

Shumai Company name: Meta Platforms, Inc.

Shumai Developer Type: company

More about Shumai, Please visit the about us page(https://facebookresearch.github.io/shumai/)

Shumai GitHub

Shumai GitHub Link: https://github.com/facebookresearch/shumai

Shumai Pricing & Plans

Open Source

Free

MIT License
Free to use, modify, and distribute
No subscription or account required

Subscribe & upgrade now

Shumai Tags

Open Source Free For Developers Linux Compatible Mac Compatible Self-Hosted

Best Suited Occupations for Shumai

Software Developers Academic Researchers Data Scientists