IntermediateИнженерияopen-source LLMs

Hugging Face LLM Course: Practical Transformers

A hands-on course on working with open-source language models via Hugging Face: from your first pipeline to deploying on Spaces and efficient fine-tuning with PEFT/LoRA.

3modules

9lessons

350 mintotal time

Developers wanting to work with open-source models via Hugging Faceaudience

Module 1

Transformers and Tokenization

Master the pipeline API, understand tokenization, and navigate the Hugging Face Hub.

Pipeline API: Run Models in Three Lines

Hugging Face pipeline() lets you run NLP tasks without knowing the model's architecture.

35 min

Tokenization: BPE, WordPiece, SentencePiece

A tokenizer splits text into subwords and converts them to numbers the model understands.

40 min

Hugging Face Hub: Finding and Downloading Models

The Hub is the main repository for open-source models, datasets, and Spaces.

30 min

Module 2

Fine-Tuning

Prepare datasets, train with the Trainer API, and efficiently fine-tune with PEFT/LoRA.

Dataset Preparation: load_dataset and Tokenization

The datasets library lets you quickly load, process, and tokenize data for training.

40 min

Trainer API: Fine-Tuning a Classifier

The Trainer class handles the training loop, logging, evaluation, and checkpoint saving.

45 min

PEFT and LoRA: Fine-Tuning Without Full Training

PEFT trains only a tiny fraction of parameters (adapters), keeping the base model frozen.

45 min

Module 3

Advanced Capabilities

Serverless inference, deploying with Gradio on Spaces, and TGI with tool-using agents.

Hugging Face Inference API: Serverless Inference

The Inference API lets you call Hub models without a GPU via HTTP or InferenceClient.

35 min

Gradio + Spaces: Deploy an ML Demo in 10 Minutes

Gradio turns a Python function into a web UI, and HF Spaces publishes it for free.

35 min

TGI + Agents: Inference Server and Tool-Using Agents

Text Generation Inference is a production LLM server with streaming and tool-call support.

45 min