7 papers
S-Agent: Spatial Tool-Use Elicits Reasoning for Spatial Intelligence
Yalun Dai, Hao Li, Shulin Tian +8
Real-world spatial intelligence requires reasoning over a continuous and evolving 3D world, yet existing VLMs and tool-augmented agents largely remain tied to static, stateless inf…
LinearARD: Linear-Memory Attention Distillation for RoPE Restoration
Ning Yang, Hengyu Zhong, Wentao Wang +4
The extension of context windows in Large Language Models is typically facilitated by scaling positional encodings followed by Continued Pre-Training (CPT). While effective, this p…
CompilerKV: Risk-Adaptive KV Compression via Offline Experience Compilation
Ning Yang, Chengzhi Wang, Yibo Liu +2
Prefill-only KV compression freezes a token subset at the end of prefill and decodes from it without further eviction. The retention decision is therefore irreversible, yet existin…
CrossCheck-Bench: Diagnosing Compositional Failures in Multimodal Conflict Resolution
Baoliang Tian, Yuxuan Si, Jilong Wang +13
Multimodal Large Language Models are primarily trained and evaluated on aligned image-text pairs, which leaves their ability to detect and resolve real-world inconsistencies largel…
Breaking the MoE LLM Trilemma: Dynamic Expert Clustering with Structured Compression
Peijun Zhu, Ning Yang, Baoliang Tian +4
Mixture-of-Experts (MoE) Large Language Models (LLMs) face a trilemma of load imbalance, parameter redundancy, and communication overhead. We introduce a unified framework based on…
Not All Errors Are Created Equal: ASCoT Addresses Late-Stage Fragility in Efficient LLM Reasoning
Dongxu Zhang, Yujun Wu, Yiding Sun +5
While Chain-of-Thought (CoT) prompting empowers Large Language Models (LLMs), ensuring reasoning reliability remains an open challenge. Contrary to the prevailing cascading failure…
Token-Importance Guided Direct Preference Optimization
Ning Yang, Hai Lin, Yibo Liu +3
Aligning Large Language Models (LLMs) with human preferences is crucial for safe and effective AI interactions. While popular methods like Direct Preference Optimization (DPO) have…