Artificial intelligence (AI) is constantly evolving, and researchers are always on the lookout for ways to improve how these systems learn. A recent breakthrough in the field has been the development of a new technique that helps AI remember old information while learning new things. This problem, known as catastrophic forgetting, has been a significant hurdle in AI development. The new method, called block expansion, has been applied to a sophisticated AI model known as the Large Language Model (LLaMA), resulting in an enhanced version dubbed LLaMA Pro.
The LLaMA 7B model, which is already quite advanced, has been upgraded with additional layers that are designed to take on new tasks without losing the knowledge it already has. This is a big step for AI systems that aim to learn continuously, much like humans do throughout their lives. The researchers behind this innovation have put the LLaMA Pro AI model to the test against various coding and math challenges. The outcome is quite remarkable: the model not only picks up new skills but also keeps up its performance on tasks it learned before. This shows that the model can handle multiple tasks effectively.
One of the key aspects of block expansion is the careful addition and specific initialization of new layers. This method ensures that the model focuses on learning new information without disrupting what it has already learned. This approach is noteworthy because it could mean that less computing power and data are needed to train large AI models, which is usually a resource-intensive process.
LLaMA Pro
“Humans generally acquire new skills without compromising the old; however, the opposite holds for Large Language Models (LLMs), e.g., from LLaMA to CodeLLaMA. To this end, we propose a new post-pretraining method for LLMs with an expansion of Transformer blocks. We tune the expanded blocks using only new corpus, efficiently and effectively improving the model’s knowledge without catastrophic forgetting. In this paper, we experiment on the corpus of code and math, yielding LLaMA Pro-8.3B, a versatile foundation model initialized from LLaMA2-7B, excelling in general tasks, programming, and mathematics.
LLaMA Pro and its instruction-following counterpart (LLaMA Pro-Instruct) achieve advanced performance among various benchmarks, demonstrating superiority over existing open models in the LLaMA family and the immense potential of reasoning and addressing diverse tasks as an intelligent agent. Our findings provide valuable insights into integrating natural and programming languages, laying a solid foundation for developing advanced language agents that operate effectively in various environments.”
Here are some other articles you may find of interest on the subject of LLaMA AI models :
The team behind this research put the LLaMA Pro model through extensive testing, which involved training it for thousands of hours on a dataset that included coding and math problems. The tests proved that the model is not only capable of taking on new challenges but also does not forget its previous training.
This advancement in the LLaMA Pro model, with its block expansion technique, represents a significant step forward in the field of machine learning. It addresses the issue of catastrophic forgetting, making AI systems more efficient and effective. As AI becomes more complex, innovations like this are crucial for the development of technology that will impact our future. Read more about the latest AI technologies in the LLaMA Pro: Progressive LLaMA with Block Expansion research paper.
Filed Under: Technology News, Top News
Latest togetherbe Deals
Disclosure: Some of our articles include affiliate links. If you buy something through one of these links, togetherbe may earn an affiliate commission. Learn about our Disclosure Policy.
