SustaiNLP 2020 - Energy Efficient NLP competition

Sharing this the SustaiNLP 2020 compeition in case anyone is interested:

Goal
Reach the lowest energy consumption for inference on SuperGLUE with performance above BERT-base or RoBERTa-large

Looks like a brilliant opportunity to revisit ULMFit/MULTiFit style models which are less computationally intensive that the large transformer models!

6 Likes

Thanks for sharing! These kind of competitions look a lot different from one that is hosted using Kaggle

1 Like

Yep typically ML conferences also tend to have some form of competition/shared task as part of them, this would be one of those

1 Like

Iā€™m up for it. @morgan