This page implements a character-level tokenization transformer that has been trained on the dataset of a popular trading card game (Yu-gi-oh! TCG). You can find the model here
.
The trained model was exported in onnx, and is loaded and used in this page via onnxruntime-web. The model thus runs on your device.
The temperature parameter rescales the logits before the softmax operation to change the model variability (standard).