LOOK-M: Look-Once Optimization in KV Cache for Efficient Multimodal Long-Context Inference
This addresses a critical bottleneck in memory and time efficiency for multimodal long-context inference, which is incremental as it builds on KV cache optimizations but targets the novel multimodal setting.
The paper tackles the problem of inefficient inference in long-context Multimodal Large Language Models (MLLMs) due to large multimodal Key-Value (KV) cache sizes, and introduces LOOK-M, a fine-tuning-free approach that reduces KV cache memory usage by up to 80% while achieving up to 1.5x faster decoding and maintaining comparable performance.
Long-context Multimodal Large Language Models (MLLMs) demand substantial computational resources for inference as the growth of their multimodal Key-Value (KV) cache, in response to increasing input lengths, challenges memory and time efficiency. Unlike single-modality LLMs that manage only textual contexts, the KV cache of long-context MLLMs includes representations from multiple images with temporal and spatial relationships and related textual contexts. The predominance of image tokens means traditional optimizations for LLMs' KV caches are unsuitable for multimodal long-context settings, and no prior works have addressed this challenge. In this work, we introduce LOOK-M, a pioneering, fine-tuning-free approach that efficiently reduces the multimodal KV cache size while maintaining performance comparable to a full cache. We observe that during prompt prefill, the model prioritizes more textual attention over image features, and based on the multimodal interaction observation, a new proposed text-prior method is explored to compress the KV cache. Furthermore, to mitigate the degradation of image contextual information, we propose several compensatory strategies using KV pairs merging. LOOK-M demonstrates that with a significant reduction in KV Cache memory usage, such as reducing it by 80% in some cases, it not only achieves up to 1.5x faster decoding but also maintains or even enhances performance across a variety of long context multimodal tasks.