Jais (language model)

From HandWiki
Revision as of 22:49, 6 February 2024 by NBrush (talk | contribs) (update)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

Jais is an open-source large language model developed in the United Arab Emirates and launched in August 2023. It was trained on both English- and Arabic-language data.

Origin

Jais is named after Jebel Jais, the highest mountain in the United Arab Emirates.[1] It was created in collaboration between the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) in Abu Dhabi, California -based Cerebras Systems and Inception, a subsidiary of G42.[1][2][3]

Training

Jais has 13 billion parameters, with an update for 30 billion in the works as of October 2023.[3] It was trained for over 21 days by a team in Abu Dhabi on a subset of Cerebras's Condor Galaxy 1 supercomputer.[1][2]

Its training dataset consisted of Arabic and English, some containing computer code.[1][3] According to Timothy Baldwin, provost, and professor of natural language processing at MBZUAI, training the model on a diverse Arabic dataset allows it to switch between dialects.[3]

Features

Jais focuses exclusively on English and Arabic translations.[4] Additional functionality for working with images, graphs and tabular data is planned for future releases.[3]

References

External links