LLM Fine-Tuning with Hugging Face: LoRA, QLoRA, PEFT

Descripción

Welcome to Fine Tuning LLM with Hugging Face Transformers for NLP, a practical and project-based course designed to help you understand and fine-tune modern Transformer models for real-world AI applications. This course starts from the basics of Hugging Face Transformers and gradually takes you into advanced fine-tuning workflows. You will learn how pipelines work, how checkpoints and models are used, how Hugging Face datasets are loaded, and how Auto Classes simplify model loading, tokenization, training, and inference. After building a strong foundation, you will go deeper into Transformer architecture. You will understand Seq2Seq models, attention mechanism, Q, K, V vectors, scaled dot-product attention, encoder-decoder stacks, positional encoding, self-attention, masked self-attention, cross-attention, and multi-head attention. The course also covers BERT architecture in detail. You will learn how BERT processes input, how masked language modeling and next sentence prediction work, and how BERT is fine-tuned for downstream NLP tasks. Then you will move into hands-on projects where you will fine-tune Transformer models for practical use cases such as sentiment classification, fake news detection, named entity recognition, text summarization, and image classification using Vision Transformers. You will also learn knowledge distillation concepts using DistilBERT, MobileBERT, and TinyBERT. This will help you understand how smaller and faster Transformer models are created for real-world production use cases. In the advanced sections, you will learn how to fine-tune LLMs on custom datasets using PEFT, LoRA, QLoRA, and 4-bit quantization. You will fine-tune models like Phi and LLaMA-style models for custom text generation and instruction/chat-based tasks. The course also includes modern Audio LLM content using Qwen3-TTS. You will learn Qwen3-TTS architecture, voice cloning, emotion control, audio data preparation, Whisper-based transcription, supervised fine-tuning, and uploading your fine-tuned audio model to Hugging Face. By the end of this course, you will have a strong practical understanding of Hugging Face Transformers and LLM fine-tuning across NLP, vision, and audio use cases. What You Will Learn • Understand Hugging Face Transformers from basic to advanced level • Use Hugging Face pipelines for NLP, vision, and audio tasks • Understand Transformer architecture, attention, encoder, decoder, and positional encoding • Learn BERT architecture, MLM, NSP, and BERT fine-tuning workflow • Fine-tune BERT for multi-class sentiment classification • Build and deploy a Streamlit app using a fine-tuned model • Understand knowledge distillation with DistilBERT, MobileBERT, and TinyBERT • Fine-tune lightweight Transformer models for fake news detection • Fine-tune DistilBERT for Named Entity Recognition • Fine-tune T5 for custom text summarization • Fine-tune Vision Transformer for Indian food image classification • Understand PEFT, LoRA, QLoRA, and 4-bit quantization • Fine-tune LLMs on custom datasets • Fine-tune a LLaMA base model into a chat/instruction model • Understand Qwen3-TTS architecture and voice cloning • Fine-tune Qwen3-TTS on custom audio data • Upload fine-tuned models to Hugging Face

Lo que aprenderás

• Understand Hugging Face Transformers and how Transformer models power modern NLP and Generative AI applications.
• Use Hugging Face pipelines, checkpoints, datasets, tokenizers, Auto Classes, and Spaces for practical AI projects.
• Learn Transformer architecture including attention, QKV vectors, encoder-decoder blocks, and positional encoding.
• Fine-tune transformers for text classification, question answering, natural language inference, text summarization, and machine translation.
• Understand BERT architecture, masked language modeling, next sentence prediction, and BERT fine-tuning.
• Fine-tune BERT for multi-class sentiment classification and build a Streamlit app for real-time prediction.
• Fine-tune DistilBERT, MobileBERT, and TinyBERT for fake news detection and performance benchmarking.
• Fine-tune Transformer models for NER, text summarization, image classification, and custom NLP tasks.
• Learn PEFT, LoRA, QLoRA, 4-bit quantization, and fine-tune LLMs on custom datasets.
• Fine-tune LLaMA-style chat models and Qwen3-TTS audio models for voice cloning and custom speech generation.

Requisitos

• Basic Python programming knowledge is required to follow the coding projects and fine-tuning notebooks.
• Basic understanding of machine learning or deep learning will be helpful but not strictly required.
• Basic NLP knowledge is useful, but important concepts are explained step by step in the course.
• A computer with internet access is required. Google Colab or a GPU machine is recommended for training.
• No prior Hugging Face experience is needed. You will learn Transformers and fine-tuning from the basics.