T5 Grammarator

Fine-tune a T5 transformer to fix grammar mistakes and serve it through a FastAPI endpoint.


Project Domains Mentors Project Difficulty
NLP, Deep Learning, Transformers, Model Deployment, Pre-processing Rakshita Kowlikar, Prithvi Tambewagh Hard

Project Description

Turn a pre-trained T5 model into a grammar-fixing engine:

  • Clean and pair error-annotated sentences from FCE or Lang-8.
  • Fine-tune T5 with Hugging Face Transformers on Google Colab GPUs.
  • Measure quality with GLEU, precision, recall, and F1.
  • Wrap the model in a lightweight FastAPI service so users can POST text and get corrected output.
    Optional extras include batch endpoints, ONNX/quantized inference, and WandB experiment tracking.

Resources

Neural Networks and LLMs
T5 Transformer