Transformers Library by Hugging Face: A Comprehensive Tutorial

Hugging Face's transformers library is the go-to library for working with state-of-the-art AI models. It offers an easy-to-use API for applying pre-trained transformer-based models for various tasks.

Let's test your understanding:

Which of these is NOT a primary domain for transformer models?

Database ManagementNatural Language ProcessingComputer VisionSpeech Recognition

Installation and Setup

Before proceeding, let's verify the installation commands:

!pip install transformers torch datasets
# Check if transformers is installed correctly
import transformers
import torch
import datasets
 
print(f"Transformers version: {transformers.__version__}")
print(f"PyTorch version: {torch.__version__}")
print(f"Datasets version: {datasets.__version__}")

Which installation command should you use?

pip install transformers torch datasetspip install huggingfacepip install ai-transformers

• transformers: Main library for AI models.
• torch: Backend framework for deep learning (can also use TensorFlow).
• datasets: Hugging Face's library for loading datasets easily.

Pretrained Models and Model Hub

Hugging Face provides a Model Hub with thousands of pre-trained models. Models are organized into domains:

• Text (NLP): GPT, BERT, RoBERTa, DistilBERT, T5, etc.
• Vision: ViT (Vision Transformer), DeiT, BEiT.
• Audio: Wav2Vec2, Whisper, etc.

Finding a Model

You can search for models here: https://huggingface.co/models

For example:

• For text tasks: bert-base-uncased, t5-small.
• For vision tasks: google/vit-base-patch16-224.
• For audio tasks: facebook/wav2vec2-base.

Let's match these models with their primary use cases:

Continue

Working with Transformers - Text

Let's try some hands-on coding with sentiment analysis:

from transformers import pipeline
 
# Create sentiment analysis pipeline
sentiment_pipeline = pipeline("sentiment-analysis")
 
# Test with different texts
texts = [
    "Hugging Face Transformers library is amazing!",
    "This code is really hard to understand",
    "I'm not sure how I feel about this"
]
 
for text in texts:
    result = sentiment_pipeline(text)
    print(f"Text: {text}")
    print(f"Sentiment: {result[0]['label']}, Score: {result[0]['score']:.4f}\n")

What sentiment would you expect for "The transformers library makes AI accessible"? POSITIVENEGATIVENEUTRAL

Working with Transformers - Vision

The Hugging Face library also supports vision transformers for image-related tasks.

Code Example: Image Classification

To classify an image, use Vision Transformer (ViT):

from transformers import pipeline
from PIL import Image
# Load an image classification pipeline
image_classifier = pipeline("image-classification", model="google/vit-base-patch16-224")
# Load an image
image = Image.open("dog.jpg")  # Use any local image path
# Perform classification
results = image_classifier(image)
for result in results:
  print(result)

Working with Transformers - Audio

For audio tasks, Hugging Face offers models like Wav2Vec2 for speech recognition.

Code Example: Speech-to-Text

from transformers import pipeline
 
# Load a speech recognition pipeline
speech_recognizer = pipeline("automatic-speech-recognition", model="facebook/wav2vec2-base-960h")
 
# Load an audio file (must be a .wav file)
audio_file = "audio_sample.wav"
 
# Transcribe the audio
result = speech_recognizer(audio_file)
print("Transcription:", result['text'])

Match these audio tasks with their appropriate models:

Continue

Fine-Tuning for Custom Tasks

from transformers import AutoTokenizer, AutoModelForSequenceClassification
 
# Load model and tokenizer
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
model = AutoModelForSequenceClassification.from_pretrained("bert-base-uncased", num_labels=2)
 
# Example text
text = "This is a sample text for fine-tuning demonstration"
 
# Tokenize
inputs = tokenizer(text, return_tensors="pt")
print("Tokenized input:", inputs)

What's the first step in fine-tuning a model?

Prepare and preprocess your datasetStart training immediatelyDownload the largest model available

Applying Transformers to Real-World Problems

Conclusion

Test your overall understanding:

Continue