Small language model
A compact model, usually under about 10 billion parameters, that can run on a laptop, a phone or modest servers.
Also: SLM, on-device model
Small language models trade some breadth for speed, low cost and privacy, since they can run without sending data to a remote server. Microsoft's Phi series, Google's Gemma and Gemini Nano, Meta's smaller Llama models, Qwen's small sizes and Apple's on-device models are examples. Training on carefully filtered and synthetic data and distilling from larger models have made them far more capable than their size would once have suggested. Many Indian startups build on small models to serve local languages cheaply.