Pre-trained compact language models ranging from 100 million to 8 billion parameters, supporting multi-modal inputs and designed for deployment on heterogeneous hardware.

A set of tools and algorithms for model compression and quantization to optimize AI models for edge deployment.