Wu Dao 2.0: China announces a 1.75T-parameter model
In one sentence At the annual conference of BAAI (Beijing Academy of Artificial Intelligence), Wu Dao 2.0 is unveiled: a 1.75-trillion-parameter multimodal Mixture of Experts model, ten times GPT-3 and larger than Switch Transformer's 1.6 trillion.
In Beijing, BAAI (Beijing Academy of Artificial Intelligence) unveils Wu Dao 2.0, a model with 1.75 trillion parameters: ten times GPT-3 and more than the 1.6 trillion of Google's Switch Transformer, the largest announced up to that point.
It's multimodal: it works on Chinese text, English text, and images, and according to BAAI it can write texts and poems in Chinese, describe images, and generate them from a description.
The announcement is read as a sign of US-China competition on large models. A paper with details and independent verification are, however, missing.
Companies
BAAI
Tools
Wu Dao 2.0
Tags
Sources
- https://en.wikipedia.org/wiki/Wu_Dao
- https://towardsdatascience.com/gpt-3-scared-you-meet-wu-dao-2-0-a-monster-of-1-75-trillion-parameters-832cd83db484/
- https://chinai.substack.com/p/chinai-145-enlightenment-via-large
- https://venturebeat.com/ai/ai-weekly-chinas-massive-multimodal-model-highlights-ai-research-gap