Alibaba Cloud, the digital technology backbone of Alibaba Group Holding, has publicly shared two of its large language models (LLMs), as the Chinese tech giant moves to expand its influence in the field of generative artificial intelligence (AI).
The two open-source models, Qwen-7B and Qwen-7B-Chat, have each been trained on 7 billion parameters. Qwen-7B-Chat is a fine-tuned version of Qwen-7B capable of conducting humanlike conversations. They represent simpler forms of Tongyi Qianwen, the LLM that Alibaba’s cloud computing service unit launched in April.
These models’ internal mechanisms – including their codes and documentation – will be made freely accessible to scholars, researchers and commercial institutions worldwide through Alibaba Cloud’s AI model repository ModelScope and the US collaborative AI platform Hugging Face, the Hangzhou-based firm said.
Companies with fewer than 100 million monthly active users will be allowed to deploy the open-source models for commercial use free of charge, while those with more users will need to request a licence from Alibaba Cloud.
LLMs are deep-learning AI algorithms that can recognise, summarise, translate and generate content using very large data sets. The technology is used to train highly intelligent chatbots like OpenAI’s ChatGPT, which has generated immense global interest since its launch in November.
Open-sourcing LLMs is part of Alibaba’s broader efforts to “democratise” AI technology and offer support for LLM start-ups, the company said.
“We aim to promote inclusive technologies and enable more developers and small and medium-sized enterprises to reap the benefits of generative AI,” Zhou Jingren, chief technology officer of Alibaba Cloud Intelligence, said.
Alibaba joins its peers in the West that have opened their LLMs to use for free. Llama 2, the next-generation open-source LLM created by Facebook owner Meta Platforms, was added to Alibaba’s ModelScope platform last month, rendering it accessible to developers in China.
ModelScope, a so-called “model-as-a-service” platform launched in November, currently features more than 1,000 ready-to-use AI models contributed by 20 leading AI institutions, covering areas from computer vision to natural language processing.
Alibaba Cloud, which is set to be spun off from its parent next year to become a publicly listed company, has been doubling down on the development and commercialisation of generative AI amid the global frenzy around ChatGPT.
At the World Artificial Intelligence Conference in Shanghai in July, it unveiled Tongyi Wanxiang, a Midjourney-style image generating model that was designed to support developers and smaller enterprises.
Alibaba’s cloud revenue in the March quarter dropped by 2 per cent from a year ago, contributing 9 per cent of the group’s total revenues, its latest financial statement showed. The company will report its June quarter results next week.