Researchers from UC Santa Cruz, UC Davis, LuxiTech, and Soochow University have developed a new method for running AI language models more efficiently by eliminating matrix multiplication. This approach could potentially reduce the environmental impact and operational costs of AI systems. Matrix multiplication is a key component of neural network computational tasks, but the researchers have created a custom 2.7 billion parameter model without using MatMul that performs similarly to conventional large language models (LLMs). They also demonstrate running a 1.3 billion parameter model at 23.8 tokens per second on a GPU accelerated by a custom-programmed FPGA chip using 13 watts of power. The researchers believe that their approach challenges the prevailing paradigm that matrix multiplication operations are essential for building high-performing language models. They suggest that their method could make large language models more accessible, efficient, and sustainable, especially for deployment on resource-constrained hardware like smartphones. While their technique has not been peer-reviewed and has limitations in testing extremely large-scale models, the researchers project that it could potentially outperform traditional LLMs at very large scales. They call for institutions with larger resources to invest in scaling up and further developing this lightweight approach to language modeling.
Source link
Source link: https://www.fudzilla.com/news/ai/59243-boffins-come-up-with-a-better-ai-model
Boffins develop improved AI model, enhancing machine learning capabilities #AIinnovation
