Beyond Accuracy: Auditing Spatial Provenance in Visual Token Pruning for OCR-Critical MLLM Inference figure
AlphaXiv 中文概览(可滚动查看)