Across three VLMs and twelve benchmarks, this work tracks the answer-relevant information decodable at each decoder layer and measures when it becomes coupled to the final output, and distinguishes the availability of visual information from its influence on prediction.
Run-Yu Zhou, Qi Zhang, Qixun Wang et al.· arXiv.org· 2 citations· ⚡1
Large language models increasingly rely on long chain-of-thought (CoT) trajectories for complex reasoning, but autoregressive generation brings substantial memory and inference costs. Latent reasoning models (LRMs) offer a more efficient alternative by compressing intermediate reasoning into a small number of continuou...
Shao-Long Chen, Ang Li, Ming-Jie Li et al.· 0 citations
This work introduces VisDocAgentBench, a closed-corpus benchmark comparing static and agentic retrieval under a shared ranked-output contract, and motivates retrieval agents that combine modality-preserving discovery with evidence-directed verification.
Lexiang Hu, Yanzhao Zhang, Mingxin Li et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.