Vision-Language Models (VLMs) have demonstrated strong performance on tasks such as video captioning and visual question answering. However, their growing scale and video-level inputs lead to significant computational and memory overhead, posing challenges for real-time deployment on hardware accelerators. While prior work attempts to reduce redundancy via token pruning or merging, these methods typically operate at coarse granularity and incur high runtime overhead due to global token-level operations.

In this study, we propose \textbf{\textit{Focus}}, a \textit{Streaming Concentration Architecture} that efficiently accelerates VLM inference through progressive, fine-grained redundancy elimination. \textit{Focus} introduces a multilevel concentration paradigm that hierarchically compresses vision-language inputs at three levels: (1) semantic-guided token pruning based on textual prompts, (2) spatial-temporal block-level concentration using localized comparisons, and (3) vector-level redundancy removal via motion-aware matching. All concentration steps are tightly co-designed with the architecture to support streaming-friendly, on-chip execution. \textit{Focus} leverages GEMM tiling, convolution-style layout, and cross-modal attention to minimize off-chip access while enabling high throughput. Implemented as a modular unit within a systolic-array accelerator, \textit{Focus} achieves up to \textbf{5.0$\times$} reduction in computation and \textbf{4.5$\times$} reduction in memory access, significantly outperforming prior hardware baselines in both performance and energy efficiency.