Repolode
Submit
huggingface

huggingface/smollm

Hugging Face's family of compact open language (SmolLM) and vision-language (SmolVLM) models designed to run on-device.

View on GitHubWebsite
Stars
3,931
Forks
326
Open issues
61
Last commit
16 days ago

What is it?

Smol Models is a family of lightweight AI models from Hugging Face, covering text and vision. The goal is to build fully open models that stay compact enough to run on-device while keeping solid performance.

The SmolLM3 language model has 3B parameters and is trained on 11T tokens. It offers a dual-mode instruct variant that can switch between think and no-think reasoning, supports six languages (English, French, Spanish, German, Italian and Portuguese), and handles context up to 128k tokens. Public training data mixtures and training configurations are released alongside the weights.

SmolVLM is a compact multimodal model for image and text input. It handles visual QA, image description, visual storytelling, and conversations that include multiple images. The repository also contains local inference tools and shared utilities, and models load through the Hugging Face transformers library.

Key features

  • 01SmolLM3 is a 3B-parameter language model with dual think/no-think reasoning modes.
  • 02Supports six languages including English, French, Spanish, German, Italian and Portuguese.
  • 03Handles context up to 128k tokens.
  • 04SmolVLM processes images and text for visual QA and image description.
  • 05Public training data mixtures and training configs are released with the weights.
  • 06Local inference tools are included for running models on-device.

Use cases

  • Running a small chat or reasoning model locally without a cloud API.
  • Building image-aware assistants that answer questions about pictures on-device.
  • Experimenting with compact multilingual models for European languages.
  • Studying how a small model is trained with fully public data and configs.

Quick start

  1. 1

    Install the Hugging Face transformers library

  2. 2

    Load SmolLM3 and generate a reply with the Python example from the README

Things to know

  • The README does not provide a pip install command, so setup requires following the linked documentation.
  • Running SmolLM3 example code expects a GPU, with CPU as an alternative.

Topics

#llm#vision-language-model#on-device#small-language-model#multimodal#transformers#open-weights

FAQ

What is smollm?
Hugging Face's family of compact open language (SmolLM) and vision-language (SmolVLM) models designed to run on-device. Smol Models is a family of lightweight AI models from Hugging Face, covering text and vision. The goal is to build fully open models that stay compact enough to run on-device while keeping solid performance.
Is smollm free to use? What license does it have?
smollm is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is smollm still maintained?
Actively maintained. The last commit to smollm was on Oct 6, 2026.
What can I use smollm for?
Running a small chat or reasoning model locally without a cloud API. Building image-aware assistants that answer questions about pictures on-device. Experimenting with compact multilingual models for European languages. Studying how a small model is trained with fully public data and configs.
How do I install smollm?
Install the Hugging Face transformers library Load SmolLM3 and generate a reply with the Python example from the README
Where can I get smollm?
The source code and installation instructions are on GitHub: https://github.com/huggingface/smollm

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects