T5-PoemGen-LoRA-PEFT

This repository contains an implementation of T5 fine-tuning using LoRA (PEFT) for instruction-based poem generation. This project focuses on resource-efficient training and detailed metric evaluation.

🚀 Quick Usage

To generate poems interactively after training, use the dedicated inference script:

python examples/inference_script.py

## ⚙️ Technical Details

This script utilizes a custom PyTorch training loop for maximum control, incorporating key optimization techniques:
* **Base Model:** T5-Base (Hugging Face)
* **PEFT Method:** LoRA (Rank 16, Alpha 32) for parameter efficiency.
* **Learning Rate Schedule:** OneCycleLR Scheduler for optimal convergence.
* **Evaluation Metrics:** Comprehensive analysis using ROUGE-L, BLEU, and METEOR metrics.