Mail fixer

Usage

Install dependencies

pip install -r requirements.txt

Generate synthetic mails

./run.sh

Train your model

python train.py

Results

We finetuned opt-125m model. The produced output is coherent, rarely repeats, creative and conveys the same information as the original mail. You can check out the model on Huggingface

Miscellaneous

Work inspired by TinyStories and Textbooks.

We use data-baby-names and occupations datasets