Alibaba's Open-Sourced AI Models Aim to Compete with Meta's Llama

Alibaba Group, the Chinese tech and e-commerce powerhouse, disclosed the launch of two open-sourced artificial intelligence (AI) models, Qwen-7B and Qwen-7B-Chat, from its cloud computing department on August 3, according to an official press release.

These two large language models (LLMs) are scaled-down versions of the previously released Tongyi Qianwen by Alibaba in April.

Each model boasts an impressive 7 billion parameters and aims to facilitate AI integration into small and medium-sized businesses.

Alibaba highlights the versatility of Qwen-7B and Qwen-7B-Chat, making their coding, model weights, and documentation accessible to academics, researchers, and commercial institutions worldwide.

Interestingly, this marks the first instance of a Chinese tech company openly sharing LLMs as open-source projects.

Zhou Jingren, chief technology officer of Alibaba Cloud Intelligence, said:

"We aim to promote inclusive technologies and enable more developers and small and medium-sized enterprises to reap the benefits of generative AI."

However, businesses with over 100 million monthly active users will need a license to utilize them fully.

Additionally, Alibaba unveiled an update to its AnalyticDB data warehousing service on August 1, featuring a vector engine that empowers corporate clients to create custom generative AI applications swiftly.

This move comes in response to Meta's recent release of Llama 2, which was co-developed with Microsoft on July 16.

Meta boasts Llama 2 being trained with 40% more public data and capable of processing twice the contextual information compared to its predecessor.

Its largest version packs an astounding 70 billion parameters.

Similar to Alibaba's approach, Meta's Llama 2 also requires a license for companies with over 700 million monthly users.

Alibaba will report on its June quarter results next week.