Meta’s LLaMA (Large Language Model Meta AI) is an open source AI model available for government, community and academic researchers and practitioners under non-commercial license. It requires “significantly less compute” than other similar products from other vendors and is available in various specifications (7B, 13B, 33B and 65B). The smallest model LLaMA 7B has undergone training on 1 billion tokens, while the LLaMA 65B and LLaMA 33B have been trained on 1.4 billion tokens. Meta states that the performance of LLaMA is better than GPT and it can even run on a single GPU. This makes it an ideal choice for researchers and practitioners looking for an open source AI model with great performance and low compute requirements.
Meta, a leading AI company, launched its large language model LLaMA in 2020. This model was designed to generate text in 20 languages, focusing on Latin and Cyrillic. It is a multi-functional basic model, applicable to many different use cases. Meta also provided a set of criteria for evaluation of model bias and toxicity to show the limitations of the model and to support further research by researchers in this critical area. In addition, Meta also launched the large language model OPT-175B and Galactica, which was reportedly capable of writing scientific articles and solving math problems. Meta’s caution in launching its models may be beneficial, as it can help to limit or eliminate bias, toxic comments, and illusions.
