Splitting the vision encoder of an MLLM into domain-specific experts with a lightweight router yields small benchmark improvements at near-zero extra inference cost.
Glam: Efficient scaling of language models with mixture-of-experts
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Mixpert: Mitigating Multimodal Learning Conflicts with Efficient Mixture-of-Vision-Experts
Splitting the vision encoder of an MLLM into domain-specific experts with a lightweight router yields small benchmark improvements at near-zero extra inference cost.