Xinyu Wang
- Block-Diagonal LoRA for Eliminating Communication Overhead in Tensor Parallel LoRA Serving
- CymbaDiff: Structured Spatial Diffusion for Sketch-based 3D Semantic Urban Scene Generation
- Mamba Modulation: On the Length Generalization of Mamba Models
- NavBench: Probing Multimodal Large Language Models for Embodied Navigation
- OCRBench v2: An Improved Benchmark for Evaluating Large Multimodal Models on Visual Text Localization and Reasoning