Empowering the world's largest computer vision ecosystem with a unified, one-click NPU hardware standard for building the next generation of real-world AI applications.

Abstract: Quantization is a critical technique employed across various research fields for compressing deep neural networks (DNNs) to facilitate deployment within resource-limited environments. This ...

turboquant-py implements the TurboQuant and QJL vector quantization algorithms from Google Research (ICLR 2026 / AISTATS 2026). It compresses high-dimensional floating-point vectors to 1-4 bits per ...

Abstract: We investigate information-theoretic limits and design of communication under receiver quantization. Unlike most existing studies that focus on low-resolution quantization, this work is more ...

Large language models (LLMs) are increasingly being deployed on edge devices—hardware that processes data locally near the data source, such as smartphones, laptops, and robots. Running LLMs on these ...

Reducing the precision of model weights can make deep neural networks run faster in less GPU memory, while preserving model accuracy. If ever there were a salient example of a counter-intuitive ...

Quantization is a process aimed at simplifying data representation by reducing precision – the number of bits used. This process involves approximating a continuous range of values with a smaller set ...