Project motivation is described in Project_Writeup.pdf. It is inspired by the Out of Order: How important is the sequential order of words in a sentence in Natural Language Understanding tasks? paper.
I tested my code with Python 3.8.6
.
<br >
pip install -r requirements.txt
Please look at settings.py
for the current tasks and models.
CUDA_VISIBLE_DEVICES=0 python create_dev_r.py
CUDA_VISIBLE_DEVICES=0 python create_dev_s.py
CUDA_VISIBLE_DEVICES=0 python compute_average_scores.py
Presentation slides can be found here. Key insights are described in the slides.
- Fine-tune
DeBerta
model on GLUE tasks. - Run the same experiments on SuperGlue benchmark. (Expectation: Word sensitivity should be higher since those tasks are more complex compared to Glue).