DataTechNotes: How to Use Hugging Face Transformers Pipeline in Python

In this post, we'll briefly learn what the Hugging Face Transformers pipeline is, how it works, and how to apply it to common NLP tasks in Python. The tutorial covers:

What is the Transformers Pipeline?
Installation
Pipeline Task Overview
Text Classification
Text Generation
Question Answering
Named Entity Recognition
Conclusion
Source Code Listing

Let's get started.

What is the Transformers Pipeline?

The Transformers pipeline is a high-level API provided by the Hugging Face transformers library that lets you run pre-trained models for common NLP tasks with just a few lines of code. Instead of manually loading tokenizers, models, and post-processing logic, the pipeline bundles everything into a single, unified interface. Under the hood it still uses the same powerful transformer-based architectures — BERT, GPT-2, RoBERTa, T5, and many others — but exposes them through a simple, task-oriented function call.

Installation

Install the required packages using pip. The transformers library works with both PyTorch and TensorFlow backends, we'll use PyTorch in this tutorial.

 
 pip install transformers torch
 

If you also want to use the sentencepiece tokenizer (required by some models such as T5 or mBART), install it as well.

 
 pip install sentencepiece

Pipeline Task Overview

The pipeline() function accepts a task string and automatically downloads a suitable default model for that task from the Hugging Face Model Hub. You can also pass a custom model argument to use a specific checkpoint.

Task String	What It Does	Default Model Family
`text-classification`	Labels text into categories (e.g., sentiment)	DistilBERT
`text-generation`	Generates a continuation of a prompt	GPT-2
`question-answering`	Extracts an answer span from a context passage	DistilBERT (SQuAD)
`ner`	Tags named entities (persons, orgs, locations)	BERT (CoNLL-2003)
`summarization`	Condenses a long document into a short summary	BART / T5
`translation`	Translates text between languages	Helsinki-NLP / T5
`fill-mask`	Predicts a masked token in a sentence	BERT / RoBERTa

Text Classification

Text classification is one of the most common NLP tasks. The pipeline returns a predicted label and a confidence score. By default, the sentiment analysis model classifies text as POSITIVE or NEGATIVE.

 
from transformers import pipeline

# Load the text-classification pipeline
classifier = pipeline("text-classification")

texts = [
    "Hugging Face makes NLP so easy and fun!",
    "The model took forever to download and then crashed.",
    "It was an average experience, nothing special."
]

results = classifier(texts)
for text, result in zip(texts, results):
    print(f"Text   : {text}")
    print(f"Label  : {result['label']},  Score: {result['score']:.4f}")
    print()
 

Output:

    
Text   : Hugging Face makes NLP so easy and fun!
Label  : POSITIVE,  Score: 0.9998

Text   : The model took forever to download and then crashed.
Label  : NEGATIVE,  Score: 0.9997

Text   : It was an average experience, nothing special.
Label  : NEGATIVE,  Score: 0.6741
 

The pipeline batches all three texts in a single forward pass and returns a list of dictionaries. The score field represents the model's confidence in its predicted label.

Text Generation

The text-generation pipeline uses an auto-regressive model (GPT-2 by default) to continue a given prompt. You can control the length and diversity of the output via max_new_tokens and temperature.

 
from transformers import pipeline

generator = pipeline("text-generation", model="gpt2")

prompt = "Transformers in machine learning are"

output = generator(
    prompt,
    max_new_tokens=60,
    num_return_sequences=2,
    temperature=0.8,
    do_sample=True,
    truncation=True
)

for i, seq in enumerate(output):
    print(f"--- Sequence {i + 1} ---")
    print(seq["generated_text"])
    print()
 

Output:

 
--- Sequence 1 ---
Transformers in machine learning are a class of deep learning models that rely
on the self-attention mechanism to capture long-range dependencies in sequential
data. They have largely replaced recurrent neural networks in NLP tasks.

--- Sequence 2 ---
Transformers in machine learning are now being applied across vision, audio, and
protein structure prediction — well beyond their original natural language
processing roots.
 

Setting num_return_sequences=2 makes the model generate two independent continuations from the same prompt. Raising temperature above 1.0 increases randomness; lowering it below 1.0 makes output more deterministic.

Question Answering

The question-answering pipeline performs extractive QA: given a context passage and a question, it identifies the span of text inside the context that best answers the question.

 
from transformers import pipeline

qa = pipeline("question-answering")

context = """
Hugging Face is an AI company founded in 2016 that develops tools for building
machine learning applications. The company is best known for its Transformers
library, which provides thousands of pre-trained models for NLP, computer vision,
and audio tasks. Its Model Hub hosts over 500,000 public models as of 2024.
"""

questions = [
    "When was Hugging Face founded?",
    "What is Hugging Face best known for?",
    "How many public models does the Model Hub host?"
]

for q in questions:
    result = qa(question=q, context=context)
    print(f"Question : {q}")
    print(f"Answer   : {result['answer']}  (score: {result['score']:.4f})")
    print()

Output:

 
Question : When was Hugging Face founded?
Answer   : 2016  (score: 0.9921)

Question : What is Hugging Face best known for?
Answer   : Transformers library  (score: 0.7308)

Question : How many public models does the Model Hub host?
Answer   : over 500,000  (score: 0.9612)
 

The model does not generate an answer from scratch — it selects a substring from context. The score reflects how confident the model is that the selected span is the correct answer.

Named Entity Recognition

Named Entity Recognition (NER) labels tokens in a sentence with entity types such as persons (PER), organizations (ORG), and locations (LOC). Using aggregation_strategy="simple" merges sub-word tokens back into full words.

 
from transformers import pipeline

ner = pipeline("ner", aggregation_strategy="simple")

sentence = (
    "Elon Musk founded SpaceX in Hawthorne, California, "
    "and later acquired Twitter, which was rebranded as X."
)

entities = ner(sentence)

print(f"{'Entity':<20} {'Type':<8} {'Score'}")
print("-" * 42)
for ent in entities:
    print(f"{ent['word']:<20} {ent['entity_group']:<8} {ent['score']:.4f}")

Output:

 
Entity               Type     Score
------------------------------------------
Elon Musk            PER      0.9987
SpaceX               ORG      0.9961
Hawthorne            LOC      0.9943
California           LOC      0.9978
Twitter              ORG      0.9832
X                    ORG      0.8754
 

The pipeline returns a list of dictionaries, each containing the entity string (word), its type (entity_group), a confidence score, and character offsets (start, end) that pinpoint the entity in the original text.

Conclusion

In this post, we briefly learned what the Hugging Face Transformers pipeline is and how it simplifies access to pre-trained models for common NLP tasks. We covered text classification, text generation, question answering, and named entity recognition — all using the same pipeline() interface with just a few lines of code. The pipeline API is an excellent starting point for prototyping and experimentation.

Source Code Listing

from transformers import pipeline

# ── Text Classification ──────────────────────────────────────────────────────
classifier = pipeline("text-classification")
texts = [
    "Hugging Face makes NLP so easy and fun!",
    "The model took forever to download and then crashed.",
    "It was an average experience, nothing special."
]
results = classifier(texts)
for text, result in zip(texts, results):
    print(f"Text: {text}")
    print(f"Label: {result['label']},  Score: {result['score']:.4f}\n")

# ── Text Generation ──────────────────────────────────────────────────────────
generator = pipeline("text-generation", model="gpt2")
prompt = "Transformers in machine learning are"
output = generator(prompt, max_new_tokens=60, num_return_sequences=2,
                   temperature=0.8, do_sample=True, truncation=True)
for i, seq in enumerate(output):
    print(f"--- Sequence {i + 1} ---\n{seq['generated_text']}\n")

# ── Question Answering ───────────────────────────────────────────────────────
qa = pipeline("question-answering")
context = """
Hugging Face is an AI company founded in 2016 that develops tools for building
machine learning applications. The company is best known for its Transformers
library, which provides thousands of pre-trained models for NLP, computer vision,
and audio tasks. Its Model Hub hosts over 500,000 public models as of 2024.
"""
questions = [
    "When was Hugging Face founded?",
    "What is Hugging Face best known for?",
    "How many public models does the Model Hub host?"
]
for q in questions:
    result = qa(question=q, context=context)
    print(f"Question : {q}")
    print(f"Answer   : {result['answer']}  (score: {result['score']:.4f})\n")

# ── Named Entity Recognition ─────────────────────────────────────────────────
ner = pipeline("ner", aggregation_strategy="simple")
sentence = (
    "Elon Musk founded SpaceX in Hawthorne, California, "
    "and later acquired Twitter, which was rebranded as X."
)
entities = ner(sentence)
print(f"{'Entity':<20} {'Type':<8} {'Score'}")
print("-" * 42)
for ent in entities:
    print(f"{ent['word']:<20} {ent['entity_group']:<8} {ent['score']:.4f}")
 

DataTechNotes

Pages

How to Use Hugging Face Transformers Pipeline in Python

What is the Transformers Pipeline?

Installation

Pipeline Task Overview

Text Classification

Text Generation

Question Answering

Named Entity Recognition

Conclusion

Source Code Listing

No comments:

Post a Comment