CVAICLMay 31

On the Limits of Token Reduction for Efficient Unified Vision Language Training

arXiv:2606.0150363.7
AI Analysis

For researchers training unified VLMs, this work identifies a fundamental limitation of naive token reduction, highlighting the importance of preserving shared cross-task structures.

The paper studies token reduction for efficient unified vision-language model training, finding that while task-specific token dropping yields efficiency gains in isolation, it causes synergy loss under joint training, indicating the need for synergy-aware acceleration strategies.

Unified vision-language models (VLMs) integrate visual understanding and visual generation within a single autoregressive backbone, but their joint training is computationally expensive and largely overlooked from an efficiency perspective. In this work, we study the feasibility and limits of token-reduction-based acceleration for unified VLM training. Through a systematic analysis of layerwise attention allocation, we uncover a fundamental asymmetry: visual understanding exhibits substantial late-layer visual redundancy, whereas visual generation maintains persistent dependence on image tokens across depth. Guided by this observation, we design task-specific accelerators that selectively reduce image-token computation for each objective. While these methods achieve significant efficiency gains in isolated settings, we observe a consistent synergy loss under unified training -- task-specific token dropping necessitates divergent parameter pathways and eliminates the mutual performance gains typically observed in joint optimization. Our findings suggest that efficient unified modeling requires preserving shared cross-task structures, highlighting the need for synergy-aware acceleration strategies. Project page: https://chicychen.github.io/TokenReductionUnifiedVLM/.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes