FUTO Unveils On-Device Swipe Typing Model

FUTO has launched FUTO Swipe, a family of open models for on-device swipe typing that can replace closed keyboard systems.

The system has an efficient architecture made up of three models:

• A 635K parameter layout-agnostic encoder
• A 300K parameter QWERTY decoder (Layout-specific decoder)
• A 1.5M parameter context language model

Thanks to its very small model size, it can run locally on-device in milliseconds (ms). Along with the model weights, FUTO has also released the 1 million swipe dataset used for training and a C++ beam search library, enabling developers to build their own swipe typing functionality across various platforms such as VR and laptops.