Overview

This repository serves as a guide to help you get started with Hugging Face, including:

• Downloading and utilizing Hugging Face models locally via the Python Transformers library.
• Constructing an API for your LLM application using FastAPI.
• Containerizing your project with Docker.
• Deploying and running your containerized application on a Kubernetes cluster.

Hugging Face

• Major hub for open-source machine learning (ML) models.
• There are hundreds of pre-trained LLMs available on Hugging Face.
• Hugging Face (HF) also has several data sets that can be used to fine tune or train LLMs.
• Hugging Spaces used to build and deploy ML models/ apps.
• Models can be found at: huggingface.co/models
• Hugging Face Hub: To make use of models hosted in HF. You will need an access token to use it.

Language models

Text-to-Text Generation (sequence to sequence) Models

• Encoder - Decoder model
• Text-to-text generation is frequently employed for tasks such as translating English sentences into French or summarizing lengthy paragraphs.
• Examples of Text Generation models include T5 and BART, which are commonly used in question-answering, Translation, and Summarization tasks.

Text Generation (Casual LM) Models

• Decoder only model.
• Often employed for tasks such as sentence completion and generating the next lines of poetry when given a few lines as input.
• Examples of Text Generation models include the GPT family, BLOOM, and PaLM, which find applications in Chatbots, Text Completion, and content generation.

Transformers library

• Python library developed by HF that makes downloading and training ML models easy.

• You can filter models that can be used with the Transformers library.

• Not all models can be used with the Transformers library.

• Pipeline function from the Transformers library can be used to interact with the LMs.

• If you don’t provide a model name, during runtime it will automatically find a default model and download it from HF.

• Second time onwards, it will use the cached copy of the model when prompted.

• Some of the currently available pipelines are:

  • feature-extraction (get the vector representation of a text)
  • Fill-mask
  • ner (named entity recognition)
  • Question-answering
  • Sentiment-analysis
  • Summarization
  • Text-generation
  • Translation
  • Zero-shot-classification

• Pretrained models are downloaded and locally cached at: ~/.cache/huggingface/hub/

References

• https://huggingface.co/
• Transformers
• NLP course