2 Getting Started

This chapter is a hands-on primer for getting productive with Hugging Face, focusing on a clean, reproducible Python setup and the tools you’ll use throughout the book. It emphasizes working in Jupyter Notebook for interactive exploration, managing dependencies with Anaconda and conda, and taking advantage of hardware acceleration when available. You also learn how to connect to the Hugging Face Hub so you can fetch models and assets programmatically and work both online and offline.

You begin by installing Anaconda, creating and activating an isolated conda environment, and launching Jupyter from a dedicated project folder. With the environment ready, you install the transformers library and, if you have a compatible GPU, set up PyTorch with CUDA (or use Apple’s MPS on Apple Silicon) to accelerate inference and training. The chapter shows how to verify GPU availability with PyTorch and inspect device details with GPUtil, then demonstrates directing transformers pipelines to the right device and even auto-selecting between CUDA, MPS, and CPU for seamless performance.

Finally, the chapter introduces the huggingface_hub package for interacting with the Hugging Face Hub from code and the command line. You learn to download specific model files (including pinned revisions), manage repositories, and authenticate securely using access tokens via the CLI or directly in Python. With these environment, acceleration, and Hub workflows in place, you’re equipped to run examples efficiently, manage dependencies safely, and integrate models and datasets into your projects with minimal friction.

Downloading Anaconda for the three major platforms – Windows, Mac, and Linux

Launching the Anaconda Prompt in Windows

Creating a new virtual environment and installing all the required packages

The virtual environment name will prefix the prompt

The web browser will now display the Jupyter Notebook’s main page

Creating a new notebook

Selecting a kernel for your notebook

The notebook is now ready to use

Renaming your notebook

The Hugging Face Hub

Downloading a file directly from a model’s page

Selecting the file you want to download

Viewing the historical commits for a project

Copying the commit hash for a file

Signing up for a new Hugging Face account

Logging in to Hugging Face Hub from Jupyter Notebook

• The Anaconda package comes with the conda package manager which simplifies package management and environment creation. It also comes with Jupyter Notebook.
• Creating virtual environments allow you to install and manage Python packages separately from your system-wide Python installation. It's a useful tool for isolating dependencies and managing different project requirements.
• The easiest way to start Jupyter Notebook is to launch it from Terminal (or Anaconda Prompt).
• The Transformers library is primarily built on top of PyTorch, a popular deep learning framework primarily developed by FaceBook’s AI Research Lab (FAIR).
• PyTorch supports GPU, enabling smooth integration with Nvidia's CUDA (Compute Unified Device Architecture), a parallel computing platform and programming model designed for GPUs.
• The Hugging Face Hub package allows you to download files, upload files, as well as perform authentication using the CLI.

FAQ

conda create -n HuggingFaceBook python=3.11 anaconda

conda activate HuggingFaceBook

mkdir HF_Projects
cd HF_Projects

jupyter notebook

!pip install transformers

pip install transformers

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121 -U

import torch
print(torch.cuda.is_available())  # True means CUDA is available

import torch
if torch.cuda.is_available():
    print("__CUDNN VERSION:", torch.backends.cudnn.version())
    print("__Number CUDA Devices:", torch.cuda.device_count())
    print("__CUDA Device Name:", torch.cuda.get_device_name(0))
    print("__CUDA Device Total Memory [GB]:",
          torch.cuda.get_device_properties(0).total_memory/1e9)

pip install GPUtil
import GPUtil
print(GPUtil.getAvailable())  # e.g., [0]
for gpu in GPUtil.getGPUs():
    print(gpu.id, gpu.name, gpu.load, gpu.memoryUtil, gpu.temperature, gpu.memoryTotal)

from transformers import pipeline
clf = pipeline("text-classification",
               model="huaen/question_detection",
               device=0)           # or device="cuda:0"

clf = pipeline("text-classification",
               model="huaen/question_detection",
               device="mps:0")

import torch
device = "cuda" if torch.cuda.is_available() else \
         ("mps" if torch.backends.mps.is_available() else "cpu")
clf = pipeline("text-classification", model="huaen/question_detection", device=device)

print(clf.device.type)  # "cuda", "mps", or "cpu"

pip install huggingface_hub

from huggingface_hub import hf_hub_download
hf_hub_download(repo_id="google/pegasus-xsum", filename="config.json")

hf_hub_download("google/pegasus-xsum", "config.json",
                revision="a0aa5531c00f59a32a167b75130805098b046f9c")

huggingface-cli login
huggingface-cli whoami

from huggingface_hub import login
login()

pip install -U ipywidgets