Build & Deploy AI with Hugging Face | Hands-On

Explore the fundamentals of Hugging Face, including model cards, dataset cards, and space cards, with a guided user interface demonstration and environment setup.

Explore Hugging Face as the open, community-driven platform that democratizes AI with open-source models, pre-trained tools, and responsible, transparent data sets for NLP, computer vision, and generative AI.

Explore the core Hugging Face components: transformers, datasets, tokenizers, hub, and spaces, and how they enable pre-trained models, data handling, fast tokenization, model hosting, and interactive apps.

Understand model, dataset, and space cards on Hugging Face and how they document resources, usage, and limitations. See how these cards promote transparency, reproducibility, and responsible AI.

Explore the hugging face UI: profile, settings, and docs; browse models, datasets, and spaces; learn to create access tokens for API usage.

Set up your development environment with two IDEs, VS Code and Google Colab, by installing VS Code, creating a Google Colab notebook, and enabling a T4 GPU for GPU acceleration.

Explore model cards and metadata, download and use models from Hugging Face Hub, explore the Hugging Face Hub Python SDK, and preview the interface while learning authentication with access tokens.

Explore Hugging Face model cards and metadata, including descriptions, usage examples, training details, evaluation metrics, and limitations, with metadata powering search, filters, and clear model provenance.

Download and use models from the Hugging Face hub, load pretrained models with transformers, manage caches, enable offline mode, and automate with the CLI and git LFS.

Explore the Hugging Face Hub Python SDK, linking your local Python environment to the hub and enabling model download, upload, repository management, authentication, caching, and offline workflows.

Install and verify the Hugging Face Hub Python SDK, log in with an access token, download files, create private repos, list models, and use offline caching.

Learn to authenticate with the Hugging Face hub using user access tokens, create and manage tokens with roles like fine-grained, read, write, or admin, and follow token security best practices.

Explore the transformers library and its architecture, covering models, tokenizers, and configuration files, then demonstrate model configurations, auto model, auto tokenizer, auto config, and customize inference pipelines.

Explore the Hugging Face Transformers library, an open source Python tool for loading and fine-tuning pre-trained models across NLP, vision, speech, and multi-modal tasks.

Explore the transformers library architecture by examining models, tokenizers, and configs; learn how tokenization converts text to input and how embeddings, encoders, and heads drive outputs.

Load and explore model configurations on hugging face, understand how configurations define architecture and tokenization behind inputs and outputs, and save, load, and customize models with auto config.

Explore auto model, auto tokenizer, and auto config in transformers, which automatically load the right model, tokenizer, and configuration for a pre-trained checkpoint, enabling simplicity and model-agnostic experimentation.

Explore how to customize inference pipelines with Hugging Face, including custom pre-processing, multi-step model chaining, batched inference, and post-processing to improve performance and enable logging.

Explore public datasets on the hugging face hub, learn dataset splits and features, tokenize data for model training, apply caching for memory efficiency, and create and upload your own dataset.

Explore public datasets on the Hugging Face Hub, understand dataset cards with metadata and tags, and load data with the datasets library to inspect splits.

Explore datasets for machine learning by loading the Iris data with pandas, examining features and labels, and creating training, validation, and test splits with scikit-learn to ensure fair, generalizable models.

Demonstrates tokenizing data for model training with Hugging Face, detailing Python-based tokenizers, converting text to input IDs and attention masks, and preparing batched, padded, truncated tensors for PyTorch.

Explore dataset caching with the Hugging Face Datasets library to cut memory usage and speed up repeated loads. The demo shows caching in VS Code and cache management with psutil.

Learn to create, structure, and upload a custom dataset with the Hugging Face datasets library, organize training and testing splits, and publish and load it from the Hugging Face hub.

Fine tune a model on a sample data set, incorporating logging, evaluation, and early stopping, then use the evaluate library to compute metrics and save and upload the fine-tuned models.

Explore how to fine-tune pre-trained models like bert, gpt-2, llama, and t5 on task-specific data to boost sentiment analysis, summarization, or question answering using Hugging Face.

Demonstrates training a sentiment classifier on a sample dataset using Hugging Face transformers in Colab, from data preparation and tokenization to training, evaluation, and saving the model.

Learn to implement logging, evaluation, and early stopping in a practical hands-on workflow using the IMDb sentiment dataset with Distilbert on Hugging Face, including dataset setup, tokenization, and trainer configuration.

Explore the evaluate library to compute and display metrics like accuracy, F1, precision, and recall for a trained model, using tokenization, datasets, and a trainer evaluation on IMDb.

learn to save and upload fine-tuned models to the hugging face hub using UI or CLI workflows. manage repositories, git lfs, and Python API uploads to deploy models efficiently.

Explore the Accelerate library and memory efficient training with gradient checkpointing, then apply hardware optimization using the Optimum library.

Discover the Hugging Face accelerate library for device-agnostic training of PyTorch models across CPU, GPU, multi-GPU, and TPU setups, with automatic distribution, mixed precision, and simple configuration.

Apply gradient checkpointing to trade memory for extra computation, enabling deeper models and larger batch sizes on GPUs with limited VRAM by recomputing activations during backpropagation.

Harness Hugging Face Optimum to optimize, quantize, and deploy transformer models across CPUs, GPUs, and accelerators, accelerating inference and reducing costs.

Explore model deployment with Hugging Face spaces, compare Gradio and Streamlit interfaces, and deploy a custom model to a Hugging Face space to enable a live interactive web interface.

Compare Gradio and Streamlit as Python frameworks for ML demos and apps, with Gradio offering quick, shareable model demos and minimal code, while Streamlit enables apps with visualization support.

Learn to deploy a custom model on Hugging Face spaces with Gradio, configure a public cpu space, upload app.py, and enable streaming chat via environment variables and token management.

Gain practical understanding to build, optimize, and deploy end-to-end AI solutions on Hugging Face using models, transformers, and spaces.