facebookresearch/MILS: Code release for "LLMs can see and hear without any training"

MILS enables large language models to handle images, audio, and video without training, performing tasks like captioning and generation across modalities, while requiring minimal computational resources and running on a single GPU.