(Yicai) Aug. 15 -- Xiaomi’s large language model, which is an artificial intelligence model that has been trained on a huge amount of language data, has achieved good results since the development team was set up in April and is helping to boost the capabilities of its virtual assistant Xiao AI, the chief executive officer of the Chinese tech giant said yesterday.

Xiaomi’s large language model is focussed on lightweight local deployment, Lei Jun, who is also founder and chairman of the Beijing-based company, said during his annual speech. It is the first time that Xiaomi has given an update on the model's progress.

The 1.3-billion-parameter large language model has been tested on mobile phones and some scenarios are comparable to a 6-billion-parameter model running in the cloud, Lei said.

Thanks to these advances, Xiao AI, which has more than 110 million monthly active users, has a stronger brain, can conduct natural dialogue, create artistic works and answer questions. Xiaomi started to invite firms to test it on Aug. 14.

Xiaomi’s previous large language model MiLM-6B had 6.4 billion parameters and achieved high accuracy in all 20 STEM subjects, including mathematics and sciences such as physics, chemistry and biology, according to the Chinese evaluation suite C-Eval.

Xiaomi will actively embrace large language models, but will not focus on general models like the US' Open AI, President Lu Weibing said at the first-quarter earnings call earlier this year. Instead, it will integrate AI into the company’s business to help improve internal efficiency.

Editor: Kim Taylor