Utilisation de LLM pré-entraînés

Introduction aux LLM en Python

Jasmin Ludolf

Senior Data Science Content Developer, DataCamp

Compréhension du langage

Person sitting at desk with several tasks around them illustrating text classification, summarization, sentiment analysis, and question and answer

Introduction aux LLM en Python

Génération de langage

Person sitting at desk with several tasks around them illustrating text generation and translation

Introduction aux LLM en Python

Génération de texte

generator = pipeline(task="text-generation", model="distilgpt2")

prompt = "The Gion neighborhood in Kyoto is famous for"

output = generator(prompt, max_length=100, pad_token_id=generator.tokenizer.eos_token_id)
  • Cohérent
  • Significatif
  • Texte similaire à celui rédigé par un être humain
  • eos_token_id : identifiant du jeton de fin de séquence
Introduction aux LLM en Python

Génération de texte

Illustration of two sequences: we should go, i really like to travel. Includes token ids and shows where padding and EOS is

  • pad_token_id : comble l'espace supplémentaire jusqu'à max_length
  • Remplissage : ajout de jetons
  • Le paramètre generator.tokenizer.eos_token_id marque la fin du texte significatif, appris lors de la formation
  • Le modèle génère jusqu'à max_length ou pad_token_id
  • truncation = True
Introduction aux LLM en Python

Génération de texte

generator = pipeline(task="text-generation", model="distilgpt2")

prompt = "The Gion neighborhood in Kyoto is famous for"

output = generator(prompt, max_length=100, pad_token_id=generator.tokenizer.eos_token_id)


print(output[0]["generated_text"])

The Gion neighborhood in Kyoto is famous for its many colorful green forests, such as the 
Red Hill, the Red River and the Red River. The Gion neighborhood is home to the world's 
tallest trees.
  • Le résultat peut être sous-optimal si le prompt est imprécis
Introduction aux LLM en Python

Orientation du résultat

generator = pipeline(task="text-generation", model="distilgpt2")


review = "This book was great. I enjoyed the plot twist in Chapter 10."

response = "Dear reader, thank you for your review."

prompt = f"Book review:

{review}
Book shop response to the review:
{response}"



output = generator(prompt, max_length=100, pad_token_id=generator.tokenizer.eos_token_id)
print(output[0]["generated_text"])

Dear reader, thank you for your review. We'd like to thank you for your reading!
Introduction aux LLM en Python

Traduction linguistique

  • Hugging Face propose une liste complète des tâches et des modèles de traduction
translator = pipeline(task="translation_en_to_es", model="Helsinki-NLP/opus-mt-en-es")


text = "Walking amid Gion's Machiya wooden houses was a mesmerizing experience."


output = translator(text, clean_up_tokenization_spaces=True)


print(output[0]["translation_text"])

Caminar entre las casas de madera Machiya de Gion fue una experiencia fascinante.
Introduction aux LLM en Python

Passons à la pratique !

Introduction aux LLM en Python

Preparing Video For Download...