11 papers
Beyond Dense Adam States: Adaptive Log-Space Quantization for Memory-Efficient Optimizers
Yan Wang
Optimizer-state quantization is commonly designed for Adam's dense, parameter-aligned first- and second-moment arrays. This abstraction breaks for memory-efficient optimizers, whos…
Libra: Taming Attention Workload Skew in Long-Context LLM Training with Bounded Sequence Pool
Yan Wang, Xiulong Yuan, Kaiming Yang +16
Long-context LLM training suffers from a load-balancing problem that sequence packing does not solve. Packing samples into fixed-token sequences balances memory and linear-cost ope…
FlashMemory-DeepSeek-V4: Lightning Index Ultra-Long Context via Lookahead Sparse Attention
Yan Wang, Qifan Zhang, Jiachen Yu +12
Conventional LLMs keep the full KV cache loaded during decoding, causing a severe GPU memory bottleneck for ultra-long context serving. In this report, we propose \textbf{Lookahead…
Hierarchical Sparse Attention Done Right: Toward Infinite Context Modeling
Xiang Hu, Xinyu Wei, Hao Gu +10
Scaling modern large language models (LLMs) to long contexts is limited by the quadratic computation cost, and poor length extrapolation of dense attention. Chunk-wise sparse atten…
From Context-Aware to Conflict-Aware: Generalizing Contrastive Decoding for Knowledge Conflict in LLMs
Runze Jiang, Taiqiang Wu, Yan Wang +2
When large language models generate from retrieved or augmented contexts, conflicts between external context and parametric priors remain a central reliability bottleneck. Existing…
AlignFed: Alignment-Aware Asynchronous Federated Fine-Tuning for Large Language Models in Heterogeneous Edge Environments
Yan Wang, Ziyi Gao, Rui Wang
Large Language Models (LLMs) have significantly propelled the advancement of edge intelligence and have been widely deployed across various scenarios, including autonomous driving,…
Planning-aligned Token Compression for Long-Context Autonomous Driving
Zhixuan Liang, Yuxiao Chen, Yurong You +12
Monolithic vision-action models represent an emerging paradigm in autonomous driving. However, this architecture produces token sequences that quickly exceed real-time computationa…
Training LLM Agents for Spontaneous, Reward-Free Self-Evolution via World Knowledge Exploration
Qifan Zhang, Dongyang Ma, Tianqing Fang +5
Most agents today ``self-evolve'' by following rewards and rules defined by humans. However, this process remains fundamentally dependent on external supervision; without human gui…
The Pensieve Paradigm: Stateful Language Models Mastering Their Own Context
Xiaoyuan Liu, Tian Liang, Dongyang Ma +4
In the world of Harry Potter, when Dumbledore's mind is overburdened, he extracts memories into a Pensieve to be revisited later. In the world of AI, while we possess the Pensieve-…
Free(): Learning to Forget in Malloc-Only Reasoning Models
Yilun Zheng, Dongyang Ma, Tian Liang +5
Reasoning models enhance problem-solving by scaling test-time compute, yet they face a critical paradox: excessive thinking tokens often degrade performance rather than improve it.…
Locas: Your Models are Principled Initializers of Locally-Supported Parametric Memories
Sidi Lu, Zhenwen Liang, Dongyang Ma +3
In this paper, we aim to bridge test-time-training with a new type of parametric memory that can be flexibly offloaded from or merged into model parameters. We present Locas, a Loc…