onnxruntime-web

A decision model that returns calibrated answers to yes/no, choice, score and multi-label questions.

An approach to make a more modular UI for ONNX Stable Diffusion, to ease future upgrades

A game where players compete to draw differing prompts on a shared canvas, as scored by a computer vision model

AlphaZero-style Quoridor engine trained from zero knowledge — no human games, no heuristics. 24 h of self-play on one consumer GPU, after which it beats the previous strongest public Quoridor AI with no search at all. Play it in your browser.

Tools for simple inference testing using CPU, WebGL, WebGPU. Simple Inference Test for onnxruntime-web.

Real-time YOLOv8m detector for 4 campus safety objects (wet-floor signs, fire alarms, exit signs, helmets). BCS407 course project at CUD; runs in the browser via ONNX Runtime Web.

Convert your openwakeword generated .onnx File to .tflite File

Small calibrated decision models (17M / 32M / 68M) for CPU and the browser. ONNX, no PyTorch at inference.

An 81M-parameter English and Roman Urdu transformer written from scratch in PyTorch, running in the browser

A 29M-parameter language model written and trained from scratch in PyTorch, running in your browser with ONNX Runtime Web.

Crop-disease inference on Kubernetes: a three-tier ladder, GPU scheduling, KEDA autoscaling on queue depth rather than CPU, and warm headroom for a cold start you cannot scale away. M1 and M2 built and measured; the on-device tier runs in the browser. The measurement says the ladder fails on field photos, and the README says so.

Paper ECG photo → digital signal → interpretation, running entirely in the browser (YOLOv8, 1D U-Net, Pyodide, ONNX Runtime Web)

Browser palm-crease segmentation: a 5.5M-param ONNX UNet that traces heart/head/life lines client-side. Live demo + full training pipeline.

Chrome extension that flags AI-generated images. All inference in-browser: no cloud, no API, no localhost backend.

Spanish text-to-speech with one-shot voice cloning, running entirely in the browser with ONNX Runtime Web. No server, no training: ported Piper + OpenVoice v2 weights.

Unofficial Talkie 1930 13B q4f16/q8 ONNX WebGPU demo with fast KV-cache browser path.

✍️ Thêm dấu tiếng Việt bằng Transformer tự viết từ đầu (PyTorch), ONNX int8 7 MB chạy ngay trên trình duyệt. word acc 0.948 · có demo

AI road damage detection, geolocation & mapping — YOLO26 / RDD-YOLO trained on RDD2022 + 9 public datasets (62.45% mAP@50). FastAPI + React + OpenStreetMap; live in-browser demo with ONNX Runtime Web.

Train a tiny LLM from scratch (PyTorch, own BPE) and chat with it fully in your browser via ONNX Runtime Web (WebGPU/WASM), zero external requests. | 从零训练小型 LLM，浏览器内本地聊天，零外部请求

걸음지기(GUARDIAN-Walk): 시각장애인 AI 청각 보행 보조 웹앱 (온디바이스 YOLO/DeepLab + GPT-4o vision, 한국어 음성)