Hi Watchers!
The superiority of smaller, specialized, more accurate AI models continues to gain attention.
Thank you Dr Sasha Luccioni and the dedicated AI sustainability folks at HuggingFace. We now have the AI Energy Score.
Developers can easily skip the enormous general purpose models and select one of the countless specialized smaller models that uses less power and gives more accurate results.
Let’s encourage everyone making key GenAI deployment decisions to take this step. Here’s an example:
For a summarization task, Google’s t5-11b consumes 44x more energy than KipperDev’s bart_summarizer_model .
(https://lnkd.in/gRWaTuFN compared to https://lnkd.in/gwcmsZ_A)
GenAI’s outsized power consumption amplifies cloud computing’s environmental impact, and it is often completely unnecessary. Let’s spread the word.
PS I am seriously fan-girling on Dr Luccioni – along with many other people!