The Chinese Academy of Sciences (CAS), China’s top scientific research institute, has released its self-developed latest-generation AI large language model called zidong taichu.
The LLM is a result of collaboration between the Institute of Automation at the CAS and the MindSpore developer community.
Developed by Huawei, MindSpore is a new open-source deep learning training/inference framework that could be used for mobile, edge and cloud scenarios.
The CAS model is an update to the 1.0 version featuring 100 billion parameters and three modalities, covering text, image and voice.
As a highlight, the new model adds modalities such as video, sensor signal and 3D point clouds.
Technically, it enables the input of structured and non-structured data in all forms of modalities and has achieved a breakthrough in core technologies like cognitive enhancement.
The LLM from CAS also supports tasks including multiple rounds of conversations with the user, textual generation, 3D comprehension and signal analytics.
It also comes with all-modality smart labeling capability, making it possible for the model to detect unidentified objects on its own.
The CAS stated that an open platform, which is developed on the basis of the zidong taichu 2.0 edition, has gone live. It is a full-stack, one-stop project constructed with low code technologies.
The platform is deployable on public or private cloud at the press of a button, thus conducive to efficient development of industry-specific LLMs.
China is caught up in a race to develop its answer to OpenAI’s ChatGPT. Over the past months, scores of Chinese companies, be they established firms or startups, have announced their bids to produce a chatbot service modelled after ChatGPT.