Small Language Model

A language model, typically ranging from tens of millions to a few billion parameters, designed to run on consumer or edge devices rather than in centralized cloud data centers. Small language models sacrifice some of the broad general knowledge of frontier large language models in exchange for low latency, predictable on-device inference, privacy (audio, text, or images never leave the device), and the ability to specialize through quantization, fine-tuning, or domain-specific lexicon injection. In accessibility contexts, SLMs underpin on-device live captioning, real-time image description, AAC prediction, and other assistive tools where cloud round-trips would be too slow or privacy-invasive.