webgpu-compute

LingPath Studio: train paired language models in the browser with CUDA kernels compiled to WebGPU.

Implementation of Reduce-then-Scan Algorithm on WebGPU

Dual-headed ResNet Go engine with MCTS, GTP, self-play training, and a WebGPU-accelerated browser demo

Browser feature test suite for compatibility with WebGPU in WASM64 and associated technologies

Browser DevTools extension for inspecting WebGL and WebGPU frames — live GPU state capture, pipeline/shader inspection, and stored-capture diffing, without replaying or reconstructing your scene.

A tiny WebGPU framework where one JS object is the single source of truth for uniforms and resources — WGSL structs, binding indices and buffer offsets are generated. Fullscreen WGSL shaders, compute passes and GPGPU in one file, no build step.

Real-time Navier-Stokes desktop wallpaper — WebGPU compute shaders (Lively Wallpaper)

An optimization idle game with an autonomous Overseer, alongside an independent Godot prototype called The Seed.

Threenix skills and reference components for Three.js developers.

High-precision 64-bit floating point math library for WebGPU

Fractals on WebGPU.

Nexus Adapter Studio is an experimental browser-based WebGPU application for training LoRA adapters on supported GGUF models. Bring your own model and JSONL corpus, evaluate learning, and export resumable Nexus adapter checkpoints. Runs locally without a Python or PyTorch backend.

PyTorch's API that trains in a browser tab on WebGPU — autograd, nn, optim, capture/replay, int8, ONNX out; held to real torch's values on 4,764 golden cases; measured on one page against TF.js, jax-js, Burn and ONNX Runtime Web (4–12× faster training, ahead of ORT at inference)

A 3D digital content creation app with webgpu

General-Purpose GPU Computing in JavaScript using WebGPU. Write GPU-accelerated code with zero boilerplate — no shader knowledge required.

The Internet is the computer. Distributed LLM inference across browsers via WebGPU.