Wang Xiaochuan, the founder of the Chinese search engine Sogou, has recently released a new language model called Baichuan-13B through his company Baichuan Intelligence. This model is restricted for commercial use by programmers and researchers. Wang Xiaochuan has stated that he wants China to have its own version of OpenAI, and Baichuan-13B is a step towards that goal.
Baichuan Intelligence launched just three months ago and quickly attracted $50 million in investments. Wang Xiaochuan’s organization is now considered one of China’s most promising creators of large language models. Baichuan-13B is built using the same Transformer design as GPT and other Chinese models. It has 13 billion parameters and is trained on data in both Chinese and English.
The model is open-source and can be used for profit. It was trained using data from GitHub. Baichuan-13B is the successor to Baichuan-7B and is a commercially available large-scale language model with 13 billion parameters. It outperforms similar models in both Chinese and English. There are two versions available: Baichuan-13B-Base, which serves as a base model for developers, and Baichuan-13B-Chat, which is more popular among regular users due to its powerful dialogue features.
Researchers are also offering int8 and int4 quantized versions of the model, which are more efficient for inference. These versions can be implemented on consumer-grade graphics cards like the Nvidia 3090. However, the non-quantized version requires more powerful hardware.
Developers can use Baichuan-13B for commercial purposes at no cost if they apply for an official commercial license. The model was trained on about 1.4 billion tokens, while ChatGPT-3 from OpenAI was trained on 300 billion tokens. The Baichuan team has grown to fifty members in just three months and they recently demonstrated their Baichuan-7B model.
Baichuan-13B is now available to researchers and programmers who have obtained the necessary legal authorization for commercial use. It’s worth noting that this model can be run on consumer hardware like Nvidia’s 3090 graphics cards, which is significant given recent U.S. restrictions against Chinese AI chip manufacturers.
Baichuan Intelligent Technology researchers have not created any Baichuan-13B-based apps for any platform yet, and they advise users not to misuse the model for illegal or harmful purposes. Users are also encouraged to follow security protocols when using the model for internet services.
In conclusion, Baichuan-13B is a highly promising language model developed by Baichuan Intelligence that has attracted significant investment and attention. It offers powerful features, outperforming similar models, and is available for commercial use. While the future of its official release for widespread use is uncertain, researchers and programmers can currently leverage its capabilities for their projects.