Continual pre-training (CPT) adapts large language models (LLMs) to specialized domains, yet its standard next-token prediction (NTP) objective treats every corpus token as a fixed target, regardless of its utility to the current model. In a controlled study on the same 1B-token mathematical corpus, we observe sharply...
Rui-Ze Zheng, Xu-Yang Liu, Siyu Ren et al.· 0 citations
Current video object removal methods suffer from prohibitive latency (minutes per video) and non-causal processing that prevents real-time interaction. We present StreamErase, enabling sub-second latency with up to 26 FPS streaming generation on a single GPU. Our approach first augments a text-to-video model with mask...
Zhuo-Bai Dong, Xiao-Bing Tu, Xian-Tao Zhang et al.· 0 citations
The results suggest that dense to MoE adaptation with dynamic expert deactivation is a practical direction for reducing active VLA model size without severe performance loss.
It is proposed that intermediate layer tokens from vision encoders should be considered for effective visual token pruning, as semantic focus shifts across layers, with middle-layer tokens capturing more detailed object-centric information that deeper layers may abstract away.
Tian-Xiang Chen, Zhentao Tan, Zi Ye et al.· IEEE Transactions on Pattern...· 0 citations
STEP-OPD is proposed, an on-policy distillation framework for image generation that extends the student's learning target beyond the teacher and introduces explicit constraints on its internal representation evolution, showing that output extrapolation enables beyond-teacher learning.
Qing-Yan Wei, Guang-Zhao Li, Xiao-Bing Tu et al.· 1 citation
This work develops HarnessBandit, an online scheduler that selects one harness per optimizer step, and improves over mixed-batch multi-harness training on both benchmarks, while training diagnostics indicate that learnability and transferability provide distinct, evolving signals.
Hong-Liang Wei, Xiao-Bing Tu, Yinggui Wang et al.· 1 citation· ⚡1
BQ-LoRA is proposed, a low-rank adaptation framework that organizes trajectory updates through a local behavior quotient manifold and contains two modules, i.e., behavior quotient balancing (BQB) and decision preserving compression (DPC).
Peng-Yang Zhou, Xiao-Bing Tu, Zheng-Xi Liu et al.· 0 citations
TAC-Merge is proposed for tracing and coordinating cross-layer influence in multimodal model merging and demonstrated the effectiveness in consolidating complementary expert capabilities and supporting generalization to unseen tasks.
Peng-Yang Zhou, Xiao-Bing Tu, Zheng-Xi Liu et al.· 0 citations
This work argues that closing this gap requires alignment at every level of the pipeline, including execution, supervision, optimization, and evaluation, rather than scale alone, and presents Wuying-Browser-Agent, a unified framework that addresses each of these levels.
A. Chen, Yan Cheng, Zhangye Han et al.· 1 citation
AViTS models spatial importance via latent-text attention and temporal importance via token-level feature variation across diffusion timesteps, and fuses them to enable spatiotemporal importance-aware selective upsampling: it prioritizes resolution refinement for critical tokens while deferring less important ones, the...
Hao-Ran Qin, Zhen Yan, Shikang Zheng et al.· 0 citations
A statistical observation that the residuals between the features of full computation steps using caching methods and reference full-compute trajectory locally exhibit a zero-mean Gaussian distribution is presented, and a plug-and-play GP-Refiner correction framework is proposed.
Zhi-Rong Shen, Rui-Xin Huang, Chang Zou et al.· 0 citations
This work proposes a general memory compression framework adaptable to various agent memory systems, MemForest, and introduces an anchor-guided propagation retrieval mechanism that retrieves relevant memory nodes from the temporal neighborhoods of key nodes, improving retrieval accuracy.
Junxi Wang, Te Sun, Jiayi Zhu 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.