Matryoshka Multimodal Models (M3) Model Card
Model details
Model type: Matryoshka Multimodal Models (M3) allow using to explicitly control visual granularities (the number of visual toknes per sample) at time time. Also, the model itself serves as a metric for image/dataset complexity. M3s is an open-source chatbot trained by fine-tuning LLaMA/Vicuna on visual conversation data. It is an auto-regressive language model, based on the transformer architecture.
Model date: llava-next-vicuna-7b-m3 was trained in May 2024. Paper
Paper or resources for more information: https://matryoshka-mm.github.io/
License
Llama 2 is licensed under the LLAMA 2 Community License, Copyright (c) Meta Platforms, Inc. All Rights Reserved.
Where to send questions or comments about the model: https://github.com/mu-cai/matryoshka-mm/issues
Intended use
Primary intended uses: The primary use of M3 is research on large multimodal models and chatbots.
Primary intended users: The primary intended users of the model are researchers and hobbyists in computer vision, natural language processing, machine learning, and artificial intelligence.
Training dataset
- 558K filtered image-text pairs from LAION/CC/SBU, captioned by BLIP.
- 665K image level instruction data from LLaVA-1.5.
Evaluation dataset
Matryoshka Multimodal Models (M3) achieves strong performance even using 1 or 9 visual tokens per image.