🔧🧠 LLM Fine-Tuning

Welcome to the LLM Fine-Tuning Repo — a collection of hands-on fine-tuning experiments on large language models (LLMs) like GPT-2(later will extend for other models too).
Here you'll find mini-projects that explore how fine-tuning transforms general-purpose models into task-specific experts. Inspired from @RajDandekar . Thanks to his LLMs from scratch playlist. I was able to create so many of these projects because of him.

🎯 What This Repo Is For

This repository is dedicated to building, training, and evaluating LLMs fine-tuned on custom datasets for various text classification and language understanding tasks.

• 🚀 Based on Transformers Architechture
• 📚 Fine-tuned from GPT-2 (for now) later will be doing on different models
• 🧪 Designed for experimentation, benchmarking, and learning

---

🧩 Current Projects

Project	Description
📘 subject-classifier	Fine-tunes GPT-2 to classify academic subjects (Math, Physics,Chemistry,Biology etc.) based on input sentences
😊 emotion-classifier	Fine-tunes GPT-2 to classify emotional tone in short texts (fear, surpise,sad, angry,joy,love etc.)
📰 news-category-classifier	Fine-tunes GPT-2 to categorize news headlines/articles into domains like Business, Sports, Technology,Science etc.

Each project has:

• ✅ A dataset that is pre-processed and is converted into suitable form
• ✅ A training script
• ✅ Evaluation metrics (accuracy)
• ✅ Sample predictions
• ✅ README for reproducibility

---

🧠 Upcoming Fine-Tuning Projects (Planned)

Here’s what I’ll be working on soon:

Project	Description
🧘 spiritual-text-classifier	Classify spiritual content (e.g., yoga, devotion, Vedanta, karma, bhakti)
📄 legal-document-tagger	Fine-tune a legal-specific classifier for document tagging
💬 multi-label tweet classifier	Handle tweets with multiple tags (sarcasm, hate speech, news, personal)
✍️ style-transfer-generator	Fine-tune an LLM to rewrite content in different tones (e.g., Shakespearean, sarcastic, formal)

---

🛠️ Tech Stack

• Transformers Architechture
• 🧨 GPT-2 as base model(later will experiment with different models)
• 🧪 Datasets: Kaggle + HuggingFace Datasets
• ⚙️ Training: full fine-tuning, LoRA (in future), PEFT(in future)

---

🚀 Getting Started

git clone https://github.com/prasanna00019/Fine-Tuning-LLMs.git
cd Fine-Tuning-LLMs
pip install -r requirements.txt