MMJamba applies a hybrid transformer-Mamba LLM backbone to multimodal instruction tuning, claiming 4x faster inference on long visual contexts and state-of-the-art accuracy on image and video benchmarks.
More specifically, we dynamically match the optimal aspect ratio from a pre-defined set of aspect ratios
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2024 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Multimodal Instruction Tuning with Hybrid State Space Models
MMJamba applies a hybrid transformer-Mamba LLM backbone to multimodal instruction tuning, claiming 4x faster inference on long visual contexts and state-of-the-art accuracy on image and video benchmarks.