SparseVILA: Decoupling Visual Sparsity for Efficient VLM Inference
This addresses scalability issues for VLM users in applications like high-resolution image understanding and long-video analysis, offering a training-free, architecture-agnostic solution, though it is incremental as it builds on existing pruning methods.
The paper tackles the problem of high inference latency in Vision Language Models (VLMs) due to many visual tokens, and presents SparseVILA, which decouples visual sparsity across prefilling and decoding stages to achieve up to 4.0x faster prefilling, 2.5x faster decoding, and 2.6x overall speedup on long-context video tasks while improving accuracy on some tasks.
Vision Language Models (VLMs) have rapidly advanced in integrating visual and textual reasoning, powering applications across high-resolution image understanding, long-video analysis, and multi-turn conversation. However, their scalability remains limited by the growing number of visual tokens that dominate inference latency. We present SparseVILA, a new paradigm for efficient VLM inference that decouples visual sparsity across the prefilling and decoding stages. SparseVILA distributes sparsity across stages by pruning redundant visual tokens during prefill and retrieving only query-relevant tokens during decoding. This decoupled design matches leading prefill pruning methods while preserving multi-turn fidelity by retaining most of the visual cache so that query-aware tokens can be retrieved at each conversation round. Built on an AWQ-optimized inference pipeline, SparseVILA achieves up to 4.0 times faster prefilling, 2.5 times faster decoding, and an overall 2.6 times end-to-end speedup on long-context video tasks -- while improving accuracy on document-understanding and reasoning tasks. By decoupling query-agnostic pruning and query-aware retrieval, SparseVILA establishes a new direction for efficient multimodal inference, offering a training-free, architecture-agnostic framework for accelerating large VLMs without sacrificing capability.