Hao Wang

Compressing Sequences in the Latent Embedding Space: $K$-Token Merging for Large Language Models

Zihao Xu, John Harvill, Ziwei Fan, et al.

IE as Cache: Information Extraction Enhanced Agentic Reasoning

Hang Lv, Sheng Liang, Hongchao Gu, et al.

Stochastic Trust-Region Methods for Over-parameterized Models

Aike Yang, Hao Wang

Visual Preference Optimization with Rubric Rewards

Ya-Qi Yu, Fangyu Hong, Xiangyang Qu, et al.

JoyAI-LLM Flash: Advancing Mid-Scale LLMs with Token Efficiency

Aichen Cai, Anmeng Zhang, Anyu Li, et al.

S2D2: Fast Decoding for Diffusion LLMs via Training-Free Self-Speculation

Ligong Han, Hao Wang, Han Gao, et al.

ImplicitRM: Unbiased Reward Modeling from Implicit Preference Data for LLM alignment

Hao Wang, Haocheng Yang, Licheng Pan, et al.

From Synthetic to Native: Benchmarking Multilingual Intent Classification in Logistics Customer Service

Haoyu He, Jinyu Zhuang, Haoran Chu, et al.

Deep Autocorrelation Modeling for Time-Series Forecasting: Progress and Prospects

Hao Wang, Licheng Pan, Qingsong Wen, et al.

Act While Thinking: Accelerating LLM Agents via Pattern-Aware Speculative Tool Execution

Yifan Sui, Han Zhao, Rui Ma, et al.

This paper addresses the slowness of AI agents (LLMs that use external tools) by proposing PASTE, a method that predicts and executes likely future tool calls before the LLM finishes thinking. By recognizing that agents tend to follow recurring patterns of tool usage, PASTE can execute tools speculatively while the LLM is still processing, hiding wait times and achieving 48.5% faster task completion.