Artificial Intelligence Laboratory PrismML developed A special version of one of its compact language models designed to work natively on smart glasses equipped with Qualcomm Snapdragon chips.
Image source: PrismML
At the Snapdragon Summit, Qualcomm demonstrated a 1-bit PrismML Bonsai model that can run natively on AI smart glasses built on the Snapdragon AR1 Gen 1 platform. PrismML works to compress large models while maintaining responsive quality – in this case, a compression rate of four times. The smart glasses version is a 2 billion parameter model optimized for processing text and visual information. In effect, owners of such glasses will be able to receive instant answers to questions about what they see before their eyes.
PrismML produces open-weight AI models designed to run directly on the device, maximizing its available computing power. The startup is positioning the solution as an alternative to relying on privacy promises from proprietary AI developers. The AI Lab, together with Qualcomm, optimized the model’s weights and architecture specifically for the Hexagon neural processor.
In terms of performance, the model is comparable to a 4-bit precision system, but consumes four times less memory and generates tokens more than twice as fast. As a result, interacting with it becomes more natural, and the context of the material remains on the device itself. PrismML and Qualcomm Technologies intend to continue working together to optimize increasingly advanced Bonsai models to run on the Snapdragon platform. Artificial intelligence technology is being introduced into smart glasses, other wearable devices, and the technology around us, so the next quantum leap cannot be achieved by simply increasing computing power.
If you find an error, select it with your mouse and press CTRL+ENTER.