ML & Transformer Notebooks

A collection of Jupyter notebooks for training purposes, designed to teach the fundamentals of transformer-based NLP models and practical machine learning workflows.

Notebooks

1. The Transformer Model (CPU)

An introduction to how transformer models process text end-to-end. Covers tokenization, inference with pre-trained models, hidden states, attention mechanisms, and sentiment analysis using DistilBERT.

2. Fine Tuning VS Full Training (GPU)

A side-by-side comparison of two training approaches on the MRPC paraphrase detection task:

• Fine-tuning a pre-trained BERT model with the HuggingFace Trainer API.
• Full training with a custom PyTorch loop including manual backpropagation, optimizer steps, and learning rate scheduling.

3. Building a Classifier Model from Scratch (GPU)

Builds a text classifier on the 20 Newsgroups dataset (18,000 posts across 20 topics) using Google Gemini API embeddings fed into a simple Keras neural network.

Disclaimer

This repository is intended solely for educational and training purposes. The notebooks, code, and examples provided here are meant to help learners understand machine learning concepts and should not be used as-is in production environments.