Natural Language Processing UT Austin Fall 2022
Final project details utilizing Transformer models:
- Adapt CheckList detailed testing approach to a industry-standard Stanford Question Answering Database (SQuAD)-trained ELECTRA model
- Train an ELECTRA SQuAD model on adversarial datasets to evaluate improvments based on question/answer categories
- Generate and train with hand-tuned training sets to improve performance on specific categories (such as a specialized model for a given task)
Models were trained and code was ran on Google CoLab Pro.
PyTorch models and HuggingFace datasets/transformers were utilized.
11-page PDF report is included under final project folder.