We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.LG

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: Multitask Prompted Training Enables Zero-Shot Task Generalization

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 held-out tasks. 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 standard datasets, often outperforming models up to 16x its size. Further, our approach attains strong performance on a subset of tasks from the BIG-bench benchmark, outperforming models up to 6x its size. All trained models are available at this https URL and all prompts are available at this https URL
Comments: ICLR 2022 Spotlight (with extended discussion)
Subjects: Machine Learning (cs.LG); Computation and Language (cs.CL)
Cite as: arXiv:2110.08207 [cs.LG]
  (or arXiv:2110.08207v3 [cs.LG] for this version)

Submission history

From: Albert Webson [view email]
[v1] Fri, 15 Oct 2021 17:08:57 GMT (7025kb,D)
[v2] Mon, 13 Dec 2021 03:31:00 GMT (7140kb,D)
[v3] Thu, 17 Mar 2022 17:53:01 GMT (7402kb,D)

Link back to: arXiv, form interface, contact.