LLaMA is a collection of language models ranging from 7B to 65B parameters

[sumaiyakhatun25]

The company said it trains its models on trillions of tokens, saying it can train advanced models using public datasets and not rely on proprietary, inaccessible datasets.

LLaMA is different.
According to Meta, training models like LLaMA requires very little computing power to test, validate, and explore new use cases. Basic language models train on large blocks of unlabeled data, making them ideal for customization for a variety of tasks.

In their research paper, Meta noted that LLaMA-13B outperformed OpenAI's GPT-3 (175B) on indonesia rcs data most benchmarks and that LLaMA-65B is competitive with the top models, Chinchilla70B by DeepMind and PaLM-540B by Google .

LLaMA is not currently used on any Meta ai products, however, the company plans to make it available to researchers. The company previously launched its LLM OPT-175B, but LLaMA is its most advanced system.

The company is making it available under a non-commercial license that focuses on research use cases. It will be available to academic researchers; those affiliated with government, civil society, and educational institutions; and industrial research laboratories worldwide.