Alibaba Open-Sources Some Generative AI Large Language Models
Chinese e-commerce giant Alibaba has open-sourced one of its large language models as well as a conversational variation of it in a bid to attract users similar to Meta’s recent Llama 2 release. The seven billion parameter version of Alibaba’s Tongyi Qianwen, known as Qwen-7B, and the Qwen-7B-Chat version for conversations are available to anyone, with a free license necessary only for companies with more than 100 million monthly users.
Open Source Qwen
Qwen-7B and Qwen-7B-Chat are both based on the Transformer neural network architecture. Alibaba claims Qwen-7b has achieved best-in-class results on several natural language processing benchmarks and that it is ahead of several potential rivals in that regard. Qwen-7B-Chat offers a specially tuned version of the model and is designed for conversational AI applications, including business needs like customer service chatbots. Alibaba says this model can engage in complex, multi-turn conversations while maintaining relevance and coherence. Both models will be publicly accessible through Alibaba Cloud.
“We aim to promote inclusive technologies and enable more developers and small and medium-sized enterprises to reap the benefits of generative AI,” Alibaba Cloud Intelligence chief technology officer Zhou Jingren said.
Alibaba launched its Tongyi Qianwen LLM in April with different parameter sizes, but making this version open-source will likely boost interest from potential customers. There’s a race among Chinese tech giants like Alibaba and Baidu to try and be China’s answer to OpenAI, though new government restrictions on generative AI may slow things down. Alibaba is pursuing several options to build its place in the Chinese LLM ecosystem. The company recently announced an agreement to support Meta’s new Llama 2 model and bring that open-source LLM to Chinese developers.
“Today, Alibaba Cloud is the first to launch a training and deployment solution for the entire series of Llama 2 in China, Alibaba posted on its WeChat channel. “All developers are welcome to create customized large models on Alibaba Cloud.”