Story
arxiv_cs_ai · Sep 9, 2026 · paper
Source brief
Beyond One-Size-Fits-All: Sample-Adaptive Strategy Routing for Vision Token Pruning in MLLMs
arxiv.orgSep 9, 2026
original source linked
In brief
Multimodal large language models (MLLMs) process hundreds or thousands of visual tokens per image, incurring prohibitive inference costs. While existing vision token pruning methods mitigate this overhead, they implic...
Feed lens
harnesseval