Repolode
Submit
huggingface

huggingface/smol-course

A hands-on Hugging Face course on aligning small language and vision-language models with minimal local hardware.

View on GitHub
Stars
6,784
Forks
2,276
Open issues
99
Last commit
16 days ago

What is it?

smol course is a practical course on aligning language models for specific use cases. It is built around the SmolLM3 and SmolVLM2 models and is designed to run on most local machines with minimal GPU requirements and no paid services.

The course covers instruction tuning with chat templates, evaluation on benchmarks and custom domains, preference alignment with methods such as DPO, and adapting vision language models. Reinforcement learning, synthetic data generation and a community leaderboard are listed as upcoming modules.

The material is open and peer reviewed, and contributors can submit exercises, examples and models through pull requests. The first version of the course, based on Markdown and Jupyter notebooks, is kept in the v1 directory.

Key features

  • 01Instruction tuning with chat templates and supervised fine-tuning
  • 02Model evaluation on standard benchmarks and custom domains
  • 03Preference alignment using algorithms such as DPO
  • 04Adapting vision language models for multimodal tasks
  • 05Runs on most local machines with minimal GPU requirements

Use cases

  • Learning how to fine-tune small LLMs for a specific domain
  • Aligning a compact model to follow instructions or human preferences on a laptop
  • Building custom evaluation sets for a niche application
  • Getting started with vision language model adaptation

Things to know

  • The README assumes prior knowledge of Python, PyTorch, and the transformers library.
  • Some course units (reinforcement learning and synthetic data) are listed as not yet released.

In collections

Learning & education tools

Topics

#llm#fine-tuning#alignment#vision-language-model#jupyter#education#hugging-face

FAQ

What is smol-course?
A hands-on Hugging Face course on aligning small language and vision-language models with minimal local hardware. smol course is a practical course on aligning language models for specific use cases. It is built around the SmolLM3 and SmolVLM2 models and is designed to run on most local machines with minimal GPU requirements and no paid services.
Is smol-course free to use? What license does it have?
smol-course is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is smol-course still maintained?
Actively maintained. The last commit to smol-course was on Oct 6, 2026.
What can I use smol-course for?
Learning how to fine-tune small LLMs for a specific domain Aligning a compact model to follow instructions or human preferences on a laptop Building custom evaluation sets for a niche application Getting started with vision language model adaptation
Where can I get smol-course?
The source code and installation instructions are on GitHub: https://github.com/huggingface/smol-course

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

Related projects