LOOM Ecosystem

GPU-accelerated neural networks in Go. Train models anywhere, deploy everywhere—browser, desktop, mobile. WebGPU compute, WASM export, multi-language bindings.

LOOM: Layered Omni-architecture Openfluke Machine

Written in Go 1.24+

Apache-2.0

WebGPU + CPU

LOOM is a modern neural network framework combining Go's simplicity with GPU acceleration. Features 5 layer types (Dense, Conv2D, Multi-Head Attention, RNN, LSTM), native Mixture of Experts, and seamless deployment across browsers (WASM), desktop, and mobile platforms with multi-language bindings.

WebGPU Compute WASM 5 Layer Types MoE (Grid Softmax) Multi-Language Cross-Platform

GitHub Repository Quick Start Guide WASM Demo Benchmarks

Core Features

LOOM provides a complete neural network framework with GPU acceleration, flexible architectures, and deployment options for any platform.

5 Layer Types

Dense, Conv2D, Multi-Head Attention, RNN, LSTM with CPU & GPU support

Grid Softmax = MoE

Native Mixture of Experts (GPT-4 architecture) - mathematically proven

WebGPU Compute

WGSL shaders for parallel execution, hybrid CPU/GPU routing

Built-in Training

Complete training loop with gradient clipping, validation, and metrics

Model Serialization

Save/load models as JSON with base64 weights - cross-platform compatible

6 Activations

ReLU, Sigmoid, Tanh, Softplus, LeakyReLU, Linear

Multi-Language Packages

Use LOOM from your favorite programming language with official bindings for Python, TypeScript/JavaScript, C#/.NET, and C/C++/Rust via C-ABI.

Native Go

Full API with WebGPU acceleration

go get github.com/openfluke/loom

Python (welvet)

High-level API via C-ABI bindings

pip install welvet

TypeScript/JavaScript

WASM module for browser deployment

npm install @openfluke/welvet

C#/.NET (Welvet)

P/Invoke bindings for .NET 9.0+

dotnet add package Welvet

WebAssembly

5.4MB binary, 24+ methods, runtime introspection

C ABI

FFI for C, C++, Rust, and other languages

Deployment Targets

Deploy LOOM neural networks anywhere: browsers via WASM, desktop applications, mobile apps, or cloud servers. Same model, same code, runs everywhere.

Browser
WebGPU / CPU
WASM (5.4MB)
Zero dependencies

Desktop
Windows
Linux
macOS

Mobile
Android
iOS
Cross-compiled

Server
Linux ARM64
x86_64
Cloud ready

Performance: GPU acceleration with WebGPU compute shaders on supported platforms. CPU fallback ensures compatibility everywhere. View benchmarks.

Open Source & Community

LOOM Resources

Main Repository

Core framework & documentation

Examples

Code samples & demos

API Docs

Detailed package documentation

Validation

98.6% loss reduction tests

Related Projects

LOOM

Open-source AI framework

IsoCard

Rendering

Portal

Web framework

View All

GitHub organization