positive
Recently
Encord launches massive open multimodal dataset and new training method
Encord’s open multimodal dataset and EBind method enable training advanced multimodal AI models on a single GPU in hours, aiming to democratise AI development for smaller teams.
Infrastructure-startup Encord has released what it calls the world’s largest open-source multimodal dataset, covering text, image, audio, video and 3D point clouds. Alongside this it introduced a new training methodology named EBind that allows multimodal models to be trained on a single GPU in hours instead of days or weeks. The company claims that by emphasising data quality and parameter efficiency, its 1.8 billion-parameter model matches the performance of models up to 17× larger.
The release is targeted especially at smaller teams and startups to democratise multimodal AI development.