7 papers
FlashBlock: Attention Caching for Efficient Long-Context Block Diffusion
Zhuokun Chen, Jianfei Cai, Bohan Zhuang
Generating long-form content, such as minute-long videos and extended texts, is increasingly important for modern generative models. Block diffusion improves inference efficiency v…
R-Stitch: Dynamic Trajectory Stitching for Efficient Reasoning
Zhuokun Chen, Zeren Chen, Jiahao He +4
Chain-of-thought (CoT) enhances the problem-solving ability of large language models (LLMs) but incurs substantial inference cost due to long autoregressive trajectories. Existing…
Where and What Matters: Sensitivity-Aware Task Vectors for Many-Shot Multimodal In-Context Learning
Ziyu Ma, Chenhui Gou, Yiming Hu +4
Large Multimodal Models (LMMs) have shown promising in-context learning (ICL) capabilities, but scaling to many-shot settings remains difficult due to limited context length and hi…
An Empirical Study on How Video-LLMs Answer Video Questions
Chenhui Gou, Ziyu Ma, Zicheng Duan +6
Taking advantage of large-scale data and pretrained language models, Video Large Language Models (Video-LLMs) have shown strong capabilities in answering video questions. However,…
MVSplat360: Feed-Forward 360 Scene Synthesis from Sparse Views
Yuedong Chen, Chuanxia Zheng, Haofei Xu +4
We introduce MVSplat360, a feed-forward approach for 360° novel view synthesis (NVS) of diverse real-world scenes, using only sparse observations. This setting is inherently ill-p…
ME-Switch: A Memory-Efficient Expert Switching Framework for Large Language Models
Jing Liu, Ruihao Gong, Mingyang Zhang +3
LLM development involves pre-training a foundation model on massive data, followed by fine-tuning on task-specific data to create specialized experts. Serving these experts can pos…