6 papers
Unveiling Downstream Performance Scaling of LLMs: A Clustering-Based Perspective
Chengyin Xu, Kaiyuan Chen, Xiao Li +2
The escalating scale and cost of Large Language Models (LLMs) training necessitate accurate pre-training prediction of downstream task performance for comprehensive understanding o…
Revealing and Enhancing Core Visual Regions: Harnessing Internal Attention Dynamics for Hallucination Mitigation in LVLMs
Guangtao Lyu, Qi Liu, Chenghao Xu +5
LVLMs have achieved strong multimodal reasoning capabilities but remain prone to hallucinations, producing outputs inconsistent with visual inputs or user instructions. Existing tr…
Theoretical Analysis of Relative Errors in Gradient Computations for Adversarial Attacks with CE Loss
Yunrui Yu, Hang Su, Cheng-zhong Xu +2
Gradient-based adversarial attacks using the Cross-Entropy (CE) loss often suffer from overestimation due to relative errors in gradient computation induced by floating-point arith…
HGCA: Hybrid GPU-CPU Attention for Long Context LLM Inference
Weishu Deng, Yujie Yang, Peiran Du +6
Scaling inference for large language models (LLMs) is increasingly constrained by limited GPU memory, especially due to growing key-value (KV) caches required for long-context gene…
Scalable Complexity Control Facilitates Reasoning Ability of LLMs
Liangkai Hang, Junjie Yao, Zhiwei Bai +17
The reasoning ability of large language models (LLMs) has been rapidly advancing in recent years, attracting interest in more fundamental approaches that can reliably enhance their…
Faster Multi-GPU Training with PPLL: A Pipeline Parallelism Framework Leveraging Local Learning
Xiuyuan Guo, Chengqi Xu, Guinan Guo +6
Currently, training large-scale deep learning models is typically achieved through parallel training across multiple GPUs. However, due to the inherent communication overhead and s…