The full name of the LLaMA language model is “Large Language Model Meta AI”, which is Meta’s new large-scale language model (LLM). parameters vary).

It is worth noting that although LaMA-13B (a model with 13 billion parameters) has more than ten times fewer model parameters than OpenAI’s GPT-3 (175 billion parameters), it can surpass the GPT-3 model in terms of performance. The smaller model also means that developers can run AI assistants like ChatGPT locally on devices such as PCs or even smartphones, without relying on large-scale facilities such as data centers.

A condensed version of LLaMA is currently available atGitHubFound on the website, Meta has no plans to publicly release the complete model and weights. Meta provides aregistration pageinterested developers can apply for registration to obtain complete codes and weights.