Poster

Multitask Prompted Training Enables Zero-Shot Task Generalization

Victor Sanh · Albert Webson · Colin Raffel · Stephen Bach · Lintang Sutawika · Zaid Alyafeai · Antoine Chaffin · Arnaud Stiegler · Arun Raja · Manan Dey · M Saiful Bari · Canwen Xu · Urmish Thakker · Shanya Sharma · Eliza Szczechla · Taewoon Kim · Gunjan Chhablani · Nihal Nayak · Debajyoti Datta · Jonathan Chang · Mike Tian-Jian Jiang · Han Wang · Matteo Manica · Sheng Shen · Zheng Xin Yong · Harshit Pandey · Rachel Bawden · Thomas Wang · Trishala Neeraj · Jos Rozen · Abheesht Sharma · Andrea Santilli · Thibault Fevry · Jason Fries · Ryan Teehan · Teven Le Scao · Stella R Biderman · Leo Gao · Thomas Wolf · Alexander M Rush


Abstract:

Large language models have recently been shown to attain reasonable zero-shot generalization on a diverse set of tasks (Brown et al., 2020). It has been hypothesized that this is a consequence of implicit multitask learning in language models’ pretraining (Radford et al., 2019). Can zero-shot generalization instead be directly induced by explicit multitask learning? To test this question at scale, we develop a system for easily mapping any natural language tasks into a human-readable prompted form. We convert a large set of supervised datasets, each with multiple prompts with diverse wording. These prompted datasets allow for benchmarking the ability of a model to perform completely unseen tasks specified in natural language. We fine-tune a pretrained encoder-decoder model (Raffel et al., 2020; Lester et al., 2021) on this multitask mixture covering a wide variety of tasks. The model attains strong zero-shot performance on several datasets, often outperforming models 16× its size. Further, our model attains strong performance on a subset of tasks from the BIG-Bench benchmark, outperforming models 6× its size. All trained models are available at https://github.com/bigscience-workshop/t-zero, and all prompts are available at https://github.com/bigscience-workshop/promptsource.

Chat is not available.