Fetching the paper…

MEDA: Dynamic KV Cache Allocation for Efficient Multimodal Long-Context Inference · Around