DyCoke: Dynamic Compression of Tokens for Fast Video Large Language Models
This addresses efficiency bottlenecks for users of VLLMs, offering a plug-and-play solution with significant speed and memory gains, though it is incremental as it builds on existing VLLM frameworks.
The paper tackles the problem of high computational cost in video large language models (VLLMs) due to thousands of visual tokens, by introducing DyCoke, a training-free token compression method that achieves 1.5X inference speedup and 1.4X memory reduction while improving performance.
Video large language models (VLLMs) have significantly advanced recently in processing complex video content, yet their inference efficiency remains constrained because of the high computational cost stemming from the thousands of visual tokens generated from the video inputs. We empirically observe that, unlike single image inputs, VLLMs typically attend visual tokens from different frames at different decoding iterations, making a one-shot pruning strategy prone to removing important tokens by mistake. Motivated by this, we present DyCoke, a training-free token compression method to optimize token representation and accelerate VLLMs. DyCoke incorporates a plug-and-play temporal compression module to minimize temporal redundancy by merging redundant tokens across frames, and applies dynamic KV cache reduction to prune spatially redundant tokens selectively. It ensures high-quality inference by dynamically retaining the critical tokens at each decoding step. Extensive experimental results demonstrate that DyCoke can outperform the prior SoTA counterparts, achieving 1.5X inference speedup, 1.4X memory reduction against the baseline VLLM, while still improving the performance, with no training.