Panoramica della libreria Transformers

Hugging Face Transformers is the go-to Python library for using and fine-tuning state-of-the-art language models with very little code. 🤗

You add the library to your project with a single pip command, and suddenly thousands of pretrained models are within reach.

Almost everything you do rests on three building blocks: a model, a tokenizer, and a high-level pipeline that ties them together.

A model holds the learned weights of a transformer like BERT or GPT. It turns input numbers into predictions such as labels or text.

A tokenizer converts your raw text into the integer ids the model expects, and converts the output back into readable text.

A pipeline bundles a tokenizer and model into one callable, so you can run a task in a single line without manual setup.

Once built, you call the pipeline like a function and it returns a clean result with a label and a confidence score.

The Hugging Face Hub hosts shared models you load by name, so you rarely need to train anything from scratch.

Helpers like AutoModel and AutoTokenizer pick the right class for any checkpoint, so one line of code works across many architectures.

Pipelines cover many jobs out of the box: classification, named-entity recognition, summarization, translation, and question answering.

Pipelines are great for trying models, but to make one truly yours you will fine-tune it on your own data later in this course.

Which piece turns raw text into the integer ids a model can read?

Transformers gives you models, tokenizers, and pipelines. Load a checkpoint from the Hub and run real NLP tasks in just a few lines. ✅