MosIT

• Paper: NExT-GPT: Any-to-Any Multimodal LLM
• GitHub Link
• Publisher: ICLR 2024
• Author Affiliation: National University of Singapore
• Type
  • SFT
  • RLHF
• Multi-turn
  • ✔
  • ✖
• Input Modalities $\rightarrow$ Output Modalities
  (I: Image, V: Video, A: Audio, 3D: Point Cloud, T: Text, B: Bounding box, Tab: Table, Web: Web page)
  • I+V+A+T $\rightarrow$ I+V+A+T
• Source
  • Youtube, Google, Flickr30k, Midjourney, etc.
• Method
  • Auto. + Manu.
• I/V/A Scale
  • I
    • 4K
  • V
    • 4K
  • A
    • 4K
• Dialog Turn
  • 4.8
• Instance Scale
  • 5K