Qwen/Qwen3-Embedding-0.6B running fully in your browser via transformers.js + ONNX WebGPU. No server, no GPU quota — inference happens on your device. · model card
The same model weights (Qwen3-Embedding-0.6B, 1024-dim) served from a 2-vCPU server on CPU.
Enter that baseline to compute the speedup. Defaults are the measured figures from
psychoguineapig on 2026-10-09.