New LLaMA AI Models Rival Giants Using Only Public Data
Meta researchers introduce LLaMA, a group of open foundation language models ranging from 7 billion to 65 billion parameters. These highly efficient models match or outperform massive proprietary systems while training exclusively on publicly available datasets.
Researchers introduce LLaMA, a collection of foundation language models ranging from 7 billion to 65 billion parameters. These models train on trillions of tokens to prove that state-of-the-art AI systems rely entirely on publicly available datasets rather than proprietary, inaccessible data.
The results show remarkable efficiency, as the smaller LLaMA-13B model outperforms the massive GPT-3 on most benchmarks. Furthermore, the largest model in the collection, LLaMA-65B, remains highly competitive with industry giants like Chinchilla-70B and PaLM-540B.
The development team releases all of these models to the broader research community. This open approach gives researchers unprecedented access to high-performing foundation models and shifts the landscape of open-source artificial intelligence development.