Meta releases LLaMA to researchers

Meta/Llama

Meta released LLaMA, a family of language models in 7B, 13B, 33B and 65B parameter sizes, trained on up to 1.4 trillion tokens. Access was granted case by case to academic, government and industry researchers under a non-commercial research licence, with the aim of letting smaller labs study large models without huge compute budgets.

• Four sizes (7B, 13B, 33B, 65B); 65B and 33B trained on 1.4 trillion tokens.
• 7B trained on 1 trillion tokens; enables research by labs without massive compute resources.
• Non-commercial research licence; access granted case-by-case to academic and industry researchers.
• Trained on text from 20 languages with most speakers, focusing on Latin and Cyrillic alphabets.