https://github.com/borhanmorphy/llm-fine-tuning-101

LLM Fine-Tuning 101
https://github.com/borhanmorphy/llm-fine-tuning-101

Last synced: 12 months ago
JSON representation

LLM Fine-Tuning 101

• Host: GitHub
• URL: https://github.com/borhanmorphy/llm-fine-tuning-101
• Owner: borhanMorphy
• License: mit
• Created: 2025-05-17T17:26:58.000Z (about 1 year ago)
• Default Branch: main
• Last Pushed: 2025-05-26T22:31:07.000Z (about 1 year ago)
• Last Synced: 2025-06-07T11:08:30.644Z (about 1 year ago)
• Language: Python
• Size: 77.1 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# 🧠 Educational Repository for Fine-Tuning Large Language Models (LLMs) — PyTorch Only

Welcome to the **LLM Fine-Tuning Educational Repository**, built entirely with **pure PyTorch** — no HuggingFace, no external frameworks. This project is designed to provide a hands-on, from-scratch learning experience for understanding and fine-tuning large language models. Perfect for learners who want to build a solid foundation by implementing everything step by step.

---

## 📚 What You'll Learn

- Core architecture and internals of LLMs and SLMs

- Dataset handling, preprocessing and multi-gpu training with PyTorch

- Building tokenizers and vocabularies from scratch

- Implementing training loops and loss functions

- Fine-tuning techniques without relying on external libraries

- LoRA (Low-Rank Adaptation) for parameter-efficient fine-tuning

- DPO (Direct Preference Optimization) for aligning models using preference data

---

## 🛠️ Repository Structure

```

.

├── demo/                 # Markdowns for demos

├── scripts/              # Utility scripts

├── slm_full_sft.py       # Supervised Full Fine-Tuning Smollm2

├── slm_lora_sft.py       # Supervised Fine-Tuning Smollm2 via LoRA

├── llm_lora_sft.py       # Supervised Fine-Tuning Gemma3 via LoRA

└── README.md             # This file

```

## 🧪 Demos

| Markdown                           | Description                             |

| ---------------------------------- | --------------------------------------- |

| `YugiohGPT.md`                     | Yugioh Card generation via LLM          |

## 💡 Why No HuggingFace?

This repo is intended for educational purposes. By not using external libraries, you’ll:

- Learn how everything works under the hood

- Gain deep insight into training dynamics and model architecture

- Build skills that translate to research and custom implementations