Hugging Face smol-course: Fine-Tuning Language Models

by Hugging Face

IntermediateCourseFree7 units, roughly 1 week each at ~3-4 hours per week

Fine-tune a small model end to end on hardware you already own — free, with a certificate.

Start LearningAdded Jul 10, 2026 · Updated Aug 11, 2026

Overview

The smol-course is Hugging Face's free, hands-on course on adapting small language models, now in its second version and delivered both on the huggingface.co/learn platform and as a GitHub repository of Jupyter notebooks. It is built around seven units: instruction tuning covering supervised fine-tuning and chat templates, evaluation with benchmarks and custom domain evals, preference alignment using algorithms such as DPO, vision language models, reinforcement learning, synthetic data generation, and a closing project showcase. The first four units are released; the reinforcement learning, synthetic data and award-ceremony units were still listed as forthcoming for October, November and December at the time of writing. Unit 1 is representative of the method: it works with SmolLM3-3B and its base checkpoint, uses the transformers library's built-in chat template support alongside TRL's SFTTrainer, walks through dataset processing and both the Python API and CLI training paths, compares base against fine-tuned behaviour, and introduces Hugging Face Jobs for training on rented cloud GPUs. Each unit pairs written material with notebooks and a quiz hosted on Hugging Face Spaces. The repository is Apache-2.0 with roughly 6.7k stars, preserves the v1 material in a separate directory, and states outright that everything runs on most local machines with minimal GPU requirements and no paid services. Two free certificates are offered: a fundamentals certificate for completing Unit 1, and a certificate of completion requiring all units plus a final project.

At a Glance

Topic: Fine-Tuning
Level: Intermediate
Format: Course
Cost: Free
Duration: 7 units, roughly 1 week each at ~3-4 hours per week
Provider: Hugging Face
Hands-on: Yes — code/exercises
Certificate: Available

What You’ll Learn

✓Apply supervised fine-tuning to a small base model with TRL's SFTTrainer
✓Format conversational datasets correctly using transformers chat templates
✓Evaluate a fine-tuned model on standard benchmarks and custom domain evals
✓Align a model to human preferences using DPO-style preference algorithms
✓Adapt and prompt multimodal vision language models for your own task
✓Run training jobs on rented GPUs through the Hugging Face Jobs service
✓Compare base and fine-tuned checkpoints to judge whether tuning actually helped

Highlights

•Free end to end, including both certificates — no paid API or subscription appears anywhere in the syllabus
•Explicitly sized for consumer hardware: minimal GPU requirements and no paid services, per the course's own README
•Taught with SmolLM3-3B, small enough that you iterate on real training runs instead of reading about them
•Every unit ships runnable notebooks in a 6.7k-star Apache-2.0 repository rather than prose alone
•Rolling release, so plan around it: three of the seven units, including reinforcement learning and synthetic data, were still unreleased at the time of writing

Who It’s For

Best For

✓ML engineers fine-tuning open models for the first time
✓Developers who want capable small models instead of frontier API calls
✓Practitioners who need a structured path from SFT through preference alignment

Prerequisites

•Familiarity with Python programming and machine learning fundamentals
•Experience with PyTorch or a similar deep learning framework
•Basic understanding of transformer architecture and LLM concepts

FAQ

What is Hugging Face smol-course: Fine-Tuning Language Models?

Hugging Face's free hands-on course on adapting small language models, now in its second version. Seven units take you from supervised fine-tuning and chat templates through evaluation, preference alignment and vision language models. Every unit ships runnable notebooks sized for consumer hardware, and two free certificates are available: one for the fundamentals unit, one for completing the course with a final project.

Is Hugging Face smol-course: Fine-Tuning Language Models free?

Hugging Face smol-course: Fine-Tuning Language Models is free to access.

What level is Hugging Face smol-course: Fine-Tuning Language Models for?

Hugging Face smol-course: Fine-Tuning Language Models is aimed at a intermediate audience. Recommended background: Familiarity with Python programming and machine learning fundamentals, Experience with PyTorch or a similar deep learning framework, Basic understanding of transformer architecture and LLM concepts.

How long does Hugging Face smol-course: Fine-Tuning Language Models take?

Expect roughly 7 units, roughly 1 week each at ~3-4 hours per week. Most learners work through it at their own pace.

What will I learn from Hugging Face smol-course: Fine-Tuning Language Models?

You'll learn: Apply supervised fine-tuning to a small base model with TRL's SFTTrainer; Format conversational datasets correctly using transformers chat templates; Evaluate a fine-tuned model on standard benchmarks and custom domain evals; Align a model to human preferences using DPO-style preference algorithms; Adapt and prompt multimodal vision language models for your own task; Run training jobs on rented GPUs through the Hugging Face Jobs service; Compare base and fine-tuned checkpoints to judge whether tuning actually helped.

Topics

fine-tuningSFTDPOTRLSmolLMHugging Face

Sources

This page was written from 3 sources, 1 on domains other than huggingface.co.