7 papers
Forget Without Compromise: Nexus Sampling for Streaming KV-Cache Eviction Under Fixed Budgets
Duc Duong, Hoang Anh Duy Le, Jianwen Xie +2
Long-context and agentic LLM workloads push the KV cache past any fixed memory budget, forcing the inference stack to permanently evict tokens at every step of a continuous-inferen…
Beyond Semantic Similarity: Rethinking Retrieval for Agentic Search via Direct Corpus Interaction
Zhuofeng Li, Haoxiang Zhang, Cong Wei +16
Modern retrieval systems, whether lexical or semantic, expose a corpus through a fixed similarity interface that compresses access into a single top-k retrieval step before reasoni…
GraphDancer: Training LLMs to Explore and Reason over Graphs via Two-Stage Curriculum Post-Training
Yuyang Bai, Zhuofeng Li, Ping Nie +2
Large language models (LLMs) increasingly rely on external knowledge to improve factuality, yet many real-world knowledge sources are organized as heterogeneous graphs rather than…
Word Salad Chopper: Reasoning Models Waste A Ton Of Decoding Budget On Useless Repetitions, Self-Knowingly
Wenya Xie, Shaochen, Zhong +4
Large Reasoning Models (LRMs) are often bottlenecked by the high cost of output tokens. We show that a significant portion of these tokens are useless self-repetitions - what we ca…
OAT-Rephrase: Optimization-Aware Training Data Rephrasing for Zeroth-Order LLM Fine-Tuning
Jikai Long, Zijian Hu, Xiaodong Yu +2
Fine-tuning large language models (LLMs) using zeroth-order optimization (ZO) offers a memory-efficient alternative to gradient-based methods but suffers from slower convergence an…
Mitigating Non-IID Drift in Zeroth-Order Federated LLM Fine-Tuning with Transferable Sparsity
Yide Ran, Wentao Guo, Jingwei Sun +7
Federated Learning enables collaborative fine-tuning of Large Language Models (LLMs) across decentralized Non-Independent and Identically Distributed (Non-IID) clients, but such mo…