Model Selection සිංහලෙන් - Part 2 | How to Fine-Tune Large Language Models (LLMs)

පාඩම 02 - Encoder / decoder or hybrid මේ අතරින් අපේ classification task එකට ගැලපෙන්නේ මොකද්ද. In this video series, you’ll learn how to fine-tune an LLM from scratch. We’ll explain all the essential theoretical concepts and coding techniques you need to master before diving into fine-tuning. Follow along with our Jupyter Notebook walkthrough to learn how to preprocess data, freeze BERT parameters, handle class imbalance, and train a high-performing spam classifier. This video is perfect for machine learning enthusiasts, NLP beginners, and data scientists looking to master fine-tuning in 2025. 🔑 What You’ll Learn: What it means to fine-tune large language models and the different methods used for fine-tuning Choosing between an encoder, decoder, or encoder-decoder model — and why it matters Preparing the fine-tuning dataset and performing preprocessing Understanding Transformers and loading base pre-trained models Tokenization and embeddings explained The self-attention mechanism in detail Creating DataLoaders for LLM training Building the LLM architecture (including fully connected layers, activation functions, dropout layers, and softmax) Using the Adam optimizer Strategies to handle class imbalance How Negative Log-Likelihood (NLL) is used in classification tasks Training and validating the model Final evaluation: loss calculation, classification report, and confusion matrix How to make inferences and predictions on new, unseen text Fine-tuning BERT for text classification using PyTorch Five fine-tuning methods: feature-based, full fine-tuning, layer-wise, adapters, and gradual unfreezing Handling class imbalance with weighted loss functions Tokenizing text using BertTokenizerFast and preparing DataLoaders