Huawei, the Chinese multinational company, is set to release its own artificial intelligence chatbot called ‘Pangu Chat.’ This chatbot is based on Huawei’s large-scale learning model, which reportedly contains 100 billion parameters, making it the first Chinese pre-trained model with such a high parameter count. Pangu Chat is expected to perform at a similar level to GPT-3, although GPT-4 is reported to have over 170 trillion parameters, surpassing Huawei’s model by a significant margin.
The Pangu Chat, named after the Pangu cloud model, will be launched on July 7th at the Huawei Developer Conference (Cloud) event. However, it will initially be available for businesses, with plans to later sell it to government and enterprise customers. Huawei has invested 960 million yuan over the past three years to train the Pangu cloud model, with a substantial amount spent on 2000 Huawei Ascend 910 chipsets and 4000 GPU/TPU cards.
Moreover, Huawei has also developed the Pangu-E large model, which is the successor to the Pangu cloud model. It has been tested to have an output of nearly 1.085 trillion parameters, competing with GPT-4 and surpassing GPT-3.5 in terms of parameter count.