LLaVaOLMoBitnet

About

LLaVaOLMoBitnet1B is an innovative ternary multimodal large language model (MM-LLM) developed by Intel Labs, pioneering the integration of image and text inputs to generate coherent textual outputs. This model is a part of a larger family ranging from 1.3B to 70B parameters, allowing for flexibility in latency and accuracy trade-offs. It features a two-step training process involving pre-training for aligning features and instruction fine-tuning, contributing to its high efficiency on smaller compute footprints. As an open-source model, it offers not only the model itself but also the training scripts, promoting further research and development in AI democratization 21.