Small language models trade some breadth or peak capability for lower memory use, latency and operating cost. They can be effective for extraction, classification, summarization and domain-specific tasks when evaluation shows that a larger model is unnecessary.
The term is relative rather than tied to one parameter threshold. A deployment should choose the smallest model that reliably meets the task's quality and safety requirements, with routing or fallback for harder requests.
ELI5
A small language model is designed to use less memory, time, and computing power than larger models. It may have less broad capability but can work well on focused tasks.
For example, a compact model can classify support messages on a local server quickly and cheaply. Harder or unusual requests can be routed to a larger model when tests show that the small model is not reliable enough.
What does SLM stand for?
SLM stands for small language model, a comparatively compact model designed for efficient language tasks.
Why use a small language model?
It can reduce latency, memory use, energy and cost for tasks that do not require a larger model's full capability.
