LLM Evaluation Notebook

This notebook demonstrates how to evaluate text outputs from large language models (LLMs) using both classical and embedding-based metrics.

---

Overview

LLM evaluation can be quantitative when you compare generated text to reference ground truth.
This notebook includes implementations for:

Metric Type	Library	Description
BLEU	nltk.translate.bleu_score	n-gram precision metric
ROUGE	rouge_score	recall-oriented overlap metric
BERTScore	bert_score	contextual embedding similarity using a pretrained transformer

---

Features

• Input data: reference and generated text pairs (from CSV or in-notebook lists)
• Compute BLEU, ROUGE-L, and BERTScore
• Aggregate metrics (mean, std, percentile)
• Easy extension for LLM benchmarks (e.g., GPT, Claude, Gemini outputs)

---

Usage

# 1) Create and activate virtual environment
python -m venv .venv
# Windows
.venv\Scripts\activate
# macOS/Linux
source .venv/bin/activate

# 2) Install dependencies
pip install -r requirements.txt

# 3) Launch notebook
jupyter lab   # or: jupyter notebook

# 4) Git clone
git clone https://github.com/Joe-Naz01/llm_evaluate.git
cd llm_evaluate