Hugging Face is an ecosystem for discovering, running, training, and sharing machine learning models and datasets, with a strong emphasis on open-source and reproducibility.

The “core four” libraries are: transformers (models + pipelines), tokenizers (fast tokenization), datasets (data loading/processing), and huggingface_hub (Hub interaction + versioning).

The Hugging Face Hub

The Hub is a Git-backed platform for hosting Models, Datasets, and Spaces (interactive demos), plus Community features for sharing and discovery.

Key definitions

• A model is a pretrained checkpoint; a tokenizer converts raw text into tokens; a pipeline bundles preprocessing, inference, and postprocessing for a task.
• A dataset is an Arrow-backed collection of data with splits (train/validation/test).
• A checkpoint is a saved snapshot of model weights/config; inference means running a trained model on new inputs; a repo is a Git-backed Hub unit storing models/datasets/Spaces.

Model Cards and Dataset Cards

• A Model Card explains intended use, training data, evaluation, limitations/biases, and licensing.
• A Dataset Card describes data sources, schema/splits, known issues/biases, ethics, and licensing.

Use cards to assess fitness-for-purpose, risk, and reproducibility.

Where to run inference?

• Run locally for control, lower latency, and offline use (you manage hardware/dependencies).
• Use an inference provider for fast setup and scalability (trade control for network latency and usage-based costs).

Workflows

Inference workflows (transformers)

Quickstart: Run inference with a pipeline

from transformers import pipeline

# Create a pipeline by specifying a task and model ID
analyze_sentiment = pipeline(
   "sentiment-analysis",
   model="distilbert-base-uncased-finetuned-sst-2-english"
)

# Run inference on input text
analyze_sentiment("Hugging Face makes NLP workflows easy!")技术支持

Text summarization

from transformers import pipeline

# Create a summarization pipeline
summarize_text = pipeline(
   "summarization",
   model="facebook/bart-large-cnn"
)

# Summarize input text
summarize_text("Long document text goes here...")技术支持

Document question answering

from transformers import pipeline

# Create a document QA pipeline
answer_question = pipeline(
   "document-question-answering",
   model="impira/layoutlm-document-qa"
)

# Ask a question about a document image
answer_question(
   image="invoice.png",
   question="What is the invoice total?"
)
技术支持

Run inference manually

import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification

model_id = "distilbert-base-uncased-finetuned-sst-2-english"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSequenceClassification.from_pretrained(model_id)

inputs = tokenizer("Hugging Face is great", return_tensors="pt")

with torch.no_grad():
   outputs = model(**inputs)

outputs.logits.argmax(dim=-1).item()技术支持

Data processing workflows (datasets)

Load and slice datasets

from datasets import load_dataset

movie_reviews = load_dataset("imdb")

train_reviews = movie_reviews["train"]
train_reviews[0]

small_sample = train_reviews.select(range(100))技术支持

Preprocess a dataset

from datasets import load_dataset
from transformers import AutoTokenizer

dataset = load_dataset("imdb")
tokenizer = AutoTokenizer.from_pretrained("distilbert-base-uncased")

def tokenize_batch(batch):
   return tokenizer(
       batch["text"],
       truncation=True,
       padding="max_length",
       max_length=256
   )

tokenized_dataset = dataset.map(
   tokenize_batch,
   batched=True,
   remove_columns=["text"]
)
技术支持

Working with the Hub (huggingface_hub)

Save locally and reload

from transformers import AutoTokenizer, AutoModelForSequenceClassification

model_id = "distilbert-base-uncased-finetuned-sst-2-english"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSequenceClassification.from_pretrained(model_id)

tokenizer.save_pretrained("local_tokenizer")
model.save_pretrained("local_model")

AutoTokenizer.from_pretrained("local_tokenizer")
AutoModelForSequenceClassification.from_pretrained("local_model")技术支持

Log in to the Hub

from huggingface_hub import login

login()
技术支持

Upload (push) a model to the Hub

from transformers import AutoTokenizer, AutoModelForSequenceClassification

repo_id = "your-username/my-model"

tokenizer = AutoTokenizer.from_pretrained("distilbert-base-uncased")
model = AutoModelForSequenceClassification.from_pretrained("distilbert-base-uncased")

tokenizer.push_to_hub(repo_id)
model.push_to_hub(repo_id)
技术支持