Fine-tuned multiple transformer models using Low-Rank Adaptation (LoRA) to accurately classify text as human-written or AI-generated, with custom dataset engineering and thorough evaluation.
GitHub: github.com/RutanshS/ai-text-detection
Languages: Python
Frameworks: PyTorch, Hugging Face Transformers, LoRA
Details#
- Fine-tuned transformer models with LoRA for efficient adaptation to the text classification task
- Engineered a custom dataset by curating and augmenting samples from the RAID repository to ensure robust model training
- Validated performance using F1-score and AUC metrics, generating confusion matrices and attention maps for detailed error analysis