VL-Cache: Sparsity and Modality-Aware KV Cache Compression for Vision-Language Model Inference Acceleration
This addresses efficiency issues for users deploying VLMs in resource-constrained environments, offering a domain-specific optimization that is incremental over existing KV cache compression methods.
The paper tackles the challenge of accelerating Vision-Language Model inference by compressing the large Key-Value cache, proposing VL-Cache to reduce memory usage and speed up processing. It achieves comparable accuracy with only 10% of the cache, accelerating latency by up to 2.33x and decoding by up to 7.08x while cutting GPU memory footprint by 90%.
Vision-Language Models (VLMs) have demonstrated impressive performance across a versatile set of tasks. A key challenge in accelerating VLMs is storing and accessing the large Key-Value (KV) cache that encodes long visual contexts, such as images or videos. While existing KV cache compression methods are effective for Large Language Models (LLMs), directly migrating them to VLMs yields suboptimal accuracy and speedup. To bridge the gap, we propose VL-Cache, a novel KV cache compression recipe tailored for accelerating VLM inference. In this paper, we first investigate the unique sparsity pattern of VLM attention by distinguishing visual and text tokens in prefill and decoding phases. Based on these observations, we introduce a layer-adaptive sparsity-aware cache budget allocation method that effectively distributes the limited cache budget across different layers, further reducing KV cache size without compromising accuracy. Additionally, we develop a modality-aware token scoring policy to better evaluate the token importance. Empirical results on multiple benchmark datasets demonstrate that retaining only 10% of KV cache achieves accuracy comparable to that with full cache. In a speed benchmark, our method accelerates end-to-end latency of generating 100 tokens by up to 2.33x and speeds up decoding by up to 7.08x, while reducing the memory footprint of KV cache in GPU by 90%.