A small language model is a language model with relatively few parameters that can also run on a laptop or phone.
Example
A small model on your phone sorts your notes without any data leaving the device.
For pros
There is no fixed limit. Depending on the source it lies at roughly 10 to 40 billion parameters; what matters is that the model runs on ordinary hardware. Small models are fast, cheap and privacy-friendly but weaker on broad knowledge. For narrowly defined agent tasks they are often enough.
