Big Model "Android Moment": After Meta, China's top open source Big Model announces "free commercial use"

After Meta Platforms Platforms Platforms Platforms Platforms Platforms Platforms announced the release of the commercial version of LLaMA, China's top open-source large-scale model ChatGLM2-6B is also available for free commercial use, marking the "Android moment" for open-source large-scale models.

On July 14th, Zhifu AI and Tsinghua KEG Laboratory announced on their official public account that in order to better support the prosperous development of the domestic open-source large-scale model ecosystem, they have decided to fully open the weights of ChatGLM-6B and ChatGLM2-6B for academic research, and allow free commercial use after completing enterprise registration and obtaining authorization.

Since the release of ChatGLM-6B on March 14th and ChatGLM2-6B on June 25th, the download volume of these two models on Huggingface has exceeded 3 million and 1.2 million respectively.

ChatGLM by Zhifu AI is currently the most attention-grabbing open-source large-scale model in China, leading in both parameters and performance in Chinese language corpus.

According to Jianzhi Research, GLM still performs exceptionally well in key indicators such as accuracy compared to large-scale models from companies like OpenAI, Alphabet-C, and Meta Platforms Platforms Platforms Platforms Platforms Platforms Platforms. It surpasses OpenAI's PaLM, Alphabet-C's PaLM, and Meta Platforms Platforms Platforms Platforms Platforms Platforms Platforms' OPT models in MMLU, LAMBADA, and BIG-bench-lite metric tests.

Currently, the training data volume of this model is 400 billion, with half in Chinese and half in English, and it has 130 billion parameters. The training cost ranges from millions to tens of millions of RMB. As of May 1st this year, the model has received download and usage requests from over 1,000 research institutions in 69 countries.

And just the day before ChatGLM2 announced its open source, there were reports that Meta Platforms is preparing to release the commercial version of its large language model LLaMA, which will allow startups and enterprises to build custom AI software based on this technology.

In February of this year, Meta Platforms released its open source large language model LLaMA, which helps researchers develop proprietary AI software at a low cost, becoming a "catfish" in the field of AI. The commercial version of LLaMA is expected to stimulate the use of this software and indirectly benefit Meta Platforms' AI development.

As mentioned in a previous article by Wall Street News, the benefits of open source models include higher user acceptance and more input data for AI processing. The more data LLaMA has, the more powerful its capabilities become. In addition, open source models enable researchers and developers to discover and address vulnerabilities while improving technology and security.

Currently, the competition among large AI models is becoming increasingly fierce, with Microsoft-backed OpenAI and Alphabet-C's non-open source platform temporarily taking the lead . However, with the commercialization of Meta Platforms and Zhishu AI & Tsinghua KEG open source models, the competition landscape in the field of artificial intelligence will be completely transformed.