Mirasol3B: A Multimodal Autoregressive model for time-aligned and contextual modalities

Benchmark Model Rank Results
action-classification-on-kinetics-soundsMirasol3B–Top 1 Accuracy: 90.1
audio-classification-on-vggsoundMirasol3B–Top 1 Accuracy: 69.8
video-question-answering-on-activitynet-qaMirasol3B–Accuracy: 51.13
video-question-answering-on-msrvtt-qaMirasol3B–Accuracy: 50.42
video-question-answering-on-next-qaMirasol3B–Accuracy: 72