Primary sourcearXiv cs.LG
View sourceEvent ·Single source
ReToken：一个Token改进视觉语言模型的视觉检索
ReToken训练一个可学习嵌入作为检索目标，从预填充的视觉KV缓存中选取与查询相关的稀疏视觉token。它在小型图像问答数据集上训练，在Visual Haystacks基准上让Qwen3VL-8B提升13.4分，InternVL3.5提升12.4分。在LVBench上，模型零样本迁移到长视频任务，Qwen3VL-8B取得8.0分增益。由于设计轻量，训练与长视频推理可在一块H100上完成。
Current read
ReToken用一个小token当检索目标，从KV缓存里挑关键视觉token，Qwen3VL-8B在Visual Haystacks涨13.4分，单卡H100就能跑。
2 sources51° AI heatLast updated 2026/7/30 17:59:56
Evidence
2 sourcesRelated source 1Yangyi
View sourceConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.