Skip to content

SmolLM

Small language models that run on local devices

Visit official site
Category AI models

What SmolLM is

The models are described as a series of small language models in 135M, 360M and 1.7B parameter versions, trained on a curated high-quality corpus. The material states they run on local devices, lower inference cost and improve privacy, with benchmarks covering common-sense reasoning and world knowledge.

What you can do with it

  • Choose among 135M, 360M and 1.7B parameter versions
  • Run the models on local devices to cut inference cost
  • Deploy with the provided ONNX and WebGPU demos
  • Work across hardware from smartphones to laptops

Application scenarios

  • Handle speech recognition and text generation on a smartphone
  • Generate and debug code on a laptop
  • Help students with complex concepts and knowledge questions in education apps

Getting started

  • Visit the Hugging Face page and download the needed model
  • Pick a 135M, 360M or 1.7B version based on device hardware
  • Deploy the model with an ONNX or WebGPU demo
  • Do instruction tuning with a publicly licensed instruction dataset

Pros & cons

✓ What we like

  • Local device use is stated to reduce inference cost
  • Hardware support spans smartphones to laptops

! What to watch out for

  • Model choice is limited to the three listed parameter sizes
  • Instruction tuning is listed as a further deployment step

FAQ

What parameter sizes are offered?

The material lists 135M, 360M and 1.7B parameter versions.

How are the models deployed?

The guide names ONNX and WebGPU demos and instruction tuning with a public dataset.

Last reviewed: 2026-11-15

More AI education tools

View all →

How we review