- A newly designed artificial intelligence (AI) tool can create images in less than two seconds and doesn’t require expensive hardware to run.
- The lowest version of this new vehicle, known as “KOALA”, has a parameter of only 700 million, meaning it is simple enough to operate quickly without the need for expensive and energy-intensive hardware.
- The tool can run on low-cost graphics processing units (GPU) and only needs around 8GB of RAM to process requests.
A newly designed artificial intelligence (AI) tool can create images in less than two seconds and doesn’t require expensive hardware to run. South Korean scientists have used special techniques called “knowledge distillation” to reduce the size of an open-source (or publicly available) rendering model, also known as Stable Diffusion He used a technique.
The lowest version of this new model, known as “KOALA”, has only 700 million parameters, meaning it is simple enough to work quickly without the need for expensive and energy-consuming hardware. The information distillation method the team uses to develop artificial intelligence allows transferring information from a large model to a smaller model without compromising its ideal performance. One of the advantages of smaller models is that they spend less time calculating and producing answers.
The tool can run on low-cost graphics processing units (GPU) and only needs around 8 GB of RAM compared to larger models that need high-end industrial GPUs to handle the demands. The team will report their findings on December 7, 2023 on arXiv published it as an article in the database. The team also published their work in an open source artificial intelligence repository Hugging Face also made available via . This innovation is behind the newly developed artificial intelligence models, as well as the Electronics and Telecommunications Research Institute (ETRI), “KOALA”, which allows creating images based on text input, and “Ko-LLaVA”, which can answer text-based questions with images or video in three versions. He created five versions in total, including two versions.
When they tested KOALA, it generated an image based on the command “Image of an astronaut reading a book under the moon on Mars” in 1.6 seconds. According to the description OpenAI’s DALL-E 2 rendered an image based on the same prompt in 12.3 seconds, while DALL-E 3 rendered it in 13.7 seconds. The scientists now plan to integrate the technology they developed into existing rendering services, education, content production and other industries.
Compiled by: Burçin Bağatur
