ExT5: Towards Extreme Multi-Task Scaling for Transfer Learning
This work addresses the challenge of optimizing multi-task learning for NLP researchers, though it is incremental in scaling existing methods.
The paper tackles the problem of scaling multi-task pre-training in NLP by introducing ExMix, a collection of 107 supervised tasks, and shows that Ex-task scaling improves performance, with ExT5 outperforming T5 baselines on benchmarks like SuperGLUE and enhancing sample efficiency.
Despite the recent success of multi-task learning and transfer learning for natural language processing (NLP), few works have systematically studied the effect of scaling up the number of tasks during pre-training. Towards this goal, this paper introduces ExMix (Extreme Mixture): a massive collection of 107 supervised NLP tasks across diverse domains and task-families. Using ExMix, we study the effect of multi-task pre-training at the largest scale to date, and analyze co-training transfer amongst common families of tasks. Through this analysis, we show that manually curating an ideal set of tasks for multi-task pre-training is not straightforward, and that multi-task scaling can vastly improve models on its own. Finally, we propose ExT5: a model pre-trained using a multi-task objective of self-supervised span denoising and supervised ExMix. Via extensive experiments, we show that ExT5 outperforms strong T5 baselines on SuperGLUE, GEM, Rainbow, Closed-Book QA tasks, and several tasks outside of ExMix. ExT5 also significantly improves sample efficiency while pre-training.