Meta Open source large language model Llama 2, Alliance with Microsoft, Qualcomm against GPT-4?

July 19 news, according to the foreign media Theverge reported that Meta officially released the first free commercial open source large language model Llama 2 through its official website in the early morning of this morning. It was also announced that Microsoft is the preferred partner for Llama 2. At the same time, Qualcomm also announced that it is working with Meta to introduce Llama into laptops, smartphones, smart headphones and other products from 2024.

According to reports, the Llama 2 model currently supports multiple parameter scale versions such as 7 billion, 13 billion, and 70 billion. Compared to the first generation of Llama, Llama 2 has been trained with 2 trillion tokens, and its length of context training is twice that of Llama, reaching 4096. In addition, the Llama-2-chat model was trained with more than 1 million manual annotations. Due to its strong parameter scale and performance, it is considered to be the most powerful challenge "GPT-4" large model.

According to Meta, Llama 2 has 40% more training data than Llama, including publicly available online sources. Llama 2 is superior to large language models such as Falcon and MPT in reasoning, editing programs, knowledge testing, etc.

In a press release, Meta explained that its decision to open source Llama 2 is to allow enterprises, startup teams, and researchers to access more AI tools and conduct more experiments and experiments.

"We believe an open approach is the right decision for AI model development today, especially in the rapidly evolving field of generative AI," Meta says, meaning developers and researchers can test, identify and solve development problems quickly.

While opening up Llama 2, Meta also wants to improve security and transparency. To this end, Llama 2 models have gone through red team drills, or generated adversarial prompts internally and externally to facilitate model fine-tuning, and also disclosed how to evaluate and adjust the model.

The open source Llama 2 will be available to customers through the Microsoft Azure platform. Meta also says that LLama will be used by customers through Amazon AWS, Hugging Face, and other services.

It is worth mentioning that Microsoft has previously allied with OpenAI and invested a huge amount of money, and a number of AI products that have been released have been imported into GPT-4. This time with Meta, Microsoft does not seem to want to put all its eggs in one basket.