NLP Labs

A collection of Natural Language Processing labs covering transformer models, text classification, and LLM fine-tuning.

Repository Structure

NLP_Labs/
├── Lab2_Intro_to_maths_optimisation/
├── Lab3_Transformer_Introduction/
├── Lab4_Transformer_Classification/
├── Lab5_Clustering/
├── Lab6_Finetune_llama/
├── Lab7_Prompting_LangChain_Ollama/
└── README.md

Lab Descriptions

Lab 2 - Introduction to Maths Optimisation

Foundations of mathematical optimization for machine learning, covering gradient descent and optimization techniques used in neural network training.

Lab 3 - Transformer Introduction

Introduction to Hugging Face Transformers library through practical NLP tasks:

• Text Classification - Sentiment analysis with DistilBERT
• Named Entity Recognition - Identifying entities (persons, organizations, locations)
• Question Answering - Extractive QA with SQuAD-trained models
• Summarization - Abstractive summarization with DistilBART
• Translation - English to German with MarianMT
• Text Generation - Autoregressive generation with GPT-2

Lab 4 - Training Transformer Models for Text Classification

End-to-end pipeline for training transformers on emotion classification:

• Dataset loading and exploration with Hugging Face Datasets
• Tokenization with DistilBERT tokenizer
• Feature extraction using pre-trained transformer hidden states
• Training a classifier on extracted embeddings
• Fine-tuning with Hugging Face Trainer API

Lab 5 - Clustering and Topic Modeling

Text clustering and topic modeling on ArXiv NLP abstracts:

• Embeddings - SentenceTransformer (gte-small)
• Dimensionality Reduction - UMAP
• Clustering - HDBSCAN
• Topic Modeling - BERTopic with c-TF-IDF
• Representation Models - KeyBERT, MMR
• LLM Topic Labels - Groq API integration

Lab 6 - Fine-tuning Llama 3.2

Fine-tuning a large language model on a medical Q&A dataset:

• Loading Llama 3.2 with Hugging Face
• LoRA (Low-Rank Adaptation) configuration for efficient fine-tuning
• Training on medical dataset
• Model evaluation and inference

Lab 7 - Prompting with LangChain & Ollama

(Instructions pending)

Setup

Each lab folder contains:

• lab.ipynb - Jupyter notebook with code and answers
• requirements.txt - Python dependencies

To run a lab:

cd Lab<N>_<name>/
pip install -r requirements.txt
jupyter notebook lab.ipynb