LLM Foundations and Systems (Exam Prep Repo)

This repository is a structured, reproducible study project covering:

• LLM Foundations & Prompting
• Data Preparation & Fine-Tuning
• Optimization & Acceleration
• Deployment & Monitoring
• Evaluation & Responsible AI

It is designed to work on macOS Catalina (10.15) with Python 3.10.x and uv.

---

Why this repo exists

The exam topics are broad. Instead of reading notes, this repo creates evidence of understanding:

• Clear notes (Markdown) for revision
• Notebooks for hands-on experiments (prompting, tokenization, evaluation)
• Systems thinking docs (deployment, monitoring, performance)
• Reproducible setup using uv + lockfile

---

Repository Structure

llm-foundations-and-systems/
├── 00_env/
│ ├── README.md
│ └── setup.md
├── 01_llm_foundations_prompting/
│ ├── README.md
│ ├── prompting_patterns.ipynb
│ └── prompt_evaluation.md
├── 02_data_and_finetuning/
│ ├── README.md
│ ├── dataset_curation.ipynb
│ ├── tokenization_analysis.ipynb
│ └── finetuning_notes.md
├── 03_optimization_and_acceleration/
│ ├── README.md
│ ├── inference_optimizations.md
│ └── training_optimizations.md
├── 04_deployment_and_monitoring/
│ ├── README.md
│ ├── inference_pipeline.md
│ └── monitoring_and_drift.md
├── 05_evaluation_and_responsible_ai/
│ ├── README.md
│ ├── evaluation_framework.ipynb
│ └── responsible_ai.md
├── pyproject.toml
├── uv.lock
└── README.md

---

Requirements

• macOS Catalina (10.15) supported
• Python: 3.10.x
• uv installed and available in PATH

Why specific pins matter on Catalina:

• numpy<2 + pyarrow<12 avoids binary incompatibilities on older macOS.
• datasets<2.19 avoids forcing newer pyarrow versions.

---

Quickstart (recommended)

From the repo root:

uv venv --python 3.10
source .venv/bin/activate
uv pip install -e ".[notebooks,llm,data,dev]"