Baichuan-13B

2023: Public Domain Publication

On July 11, 2023 Chinesestartup , Baichuan Intelligence introduced a large artificial intelligence Baichuan-13B language model with approximately 13 billion parameters. It is considered as a local analogue of the platform. OpenAI ChatGPT

Baichuan Intelligence was founded by Chinese entrepreneur Wang Xiaochuan, the creator of the Sogou search engine (part of Tencent). According to him, "China needs its own OpenAI." Startup Baichuan Intelligence may become one of the most promising developers of large language models in the PRC market.

Baichuan Intelligence introduced a large language model Baichuan-13B

Wang retired from Sogou at the end of 2021 and amid the soaring popularity of chatbot ChatGPT has formed a new firm. By mid-July 2023, about $50 million was raised for the development of Baichuan Intelligence. The presented model Baichuan-13B built on the Transformer architecture (underpins GPT). The system is trained on Chinese and English data. The source code of the platform is posted on GitHub and is available for commercial use.

It is noted that the model Baichuan-13B trained at 1.4 trillion tokens. For comparison, LLaMa from Meta (recognized as an extremist organization; activities in the Russian Federation are prohibited) uses 1 billion tokens in a model that also has 13 billion parameters. Wang said his startup is preparing to bring a large-scale model comparable to OpenAI GPT-3.5 to market by the end of 2023.

The Baichuan-13B model is available for free to scientists and developers who have received official permission to use it for commercial purposes. In light of US sanctions, the platform allows options that can run on consumer-grade hardware, including Nvidia 3090 graphics accelerators.^[1][2]

Notes