Zhenzhu Xu

Zhichao Xu, Minheng Wang, Yawei Wang et al.

Retrieval-augmented generation (RAG) systems trained using reinforcement learning (RL) with reasoning are hampered by inefficient context management, where long, noisy retrieved documents increase costs and degrade performance. We introduce RECON (REasoning with CONdensation), a framework that integrates an explicit summarization module to compress evidence within the reasoning loop. Our summarizer is trained via a two-stage process: relevance pretraining on QA datasets, followed by multi-aspect distillation from proprietary LLMs to ensure factuality and clarity. Integrated into the Search-R1 pipeline, RECON reduces total context length by 35\%, leading to improved training speed and inference latency, while simultaneously improving RAG performance on downstream QA benchmarks. Notably, it boosts the average EM score of the 3B model by 14.5\% and the 7B model by 3.0\%, showing particular strength in multi-hop QA. RECON demonstrates that learned context compression is essential for building practical, scalable, and performant RAG systems. Our code implementation is made available at https://github.com/allfornancy/RECON.

Zhentao Xu, Fengyi Li, Albert Chen et al.

In large-scale industrial LLM systems, prompt templates often expand to thousands of tokens as teams iteratively incorporate sections such as task instructions, few-shot examples, and heuristic rules to enhance robustness and coverage. This expansion leads to bloated prompts that are difficult to maintain and incur significant inference latency and serving costs. To address this, we introduce Prompt Compression via Attribution Estimation (ProCut), a flexible, LLM-agnostic, training-free framework that compresses prompts through attribution analysis. ProCut segments prompt templates into semantically meaningful units, quantifies their impact on task performance, and prunes low-utility components. Through extensive experiments on five public benchmark datasets and real-world industrial prompts, we show that ProCut achieves substantial prompt size reductions (78% fewer tokens in production) while maintaining or even slightly improving task performance (up to 62% better than alternative methods). We further introduce an LLM-driven attribution estimator that reduces compression latency by over 50%, and demonstrate that ProCut integrates seamlessly with existing prompt-optimization frameworks to produce concise, high-performing prompts.

Zheng Huang, Kiran Ramnath, Yueyan Chen et al.

Diffusion language models (DLMs) have recently emerged as a compelling alternative to autoregressive generation, offering parallel generation and improved global coherence. During inference, DLMs generate text by iteratively denoising masked sequences in parallel; however, determining which positions to unmask and which tokens to commit forms a large combinatorial search problem. Existing inference methods approximate this search using heuristics, which often yield suboptimal decoding paths; other approaches instead rely on additional training to guide token selection. To introduce a principled search mechanism for DLMs inference, we introduce MEDAL, an inference-time scaling framework that integrates Monte Carlo Tree SEarch initialization for Diffusion LAnguage Model inference. We employ Monte Carlo Tree Search at the initialization stage to explore promising unmasking trajectories, providing a robust starting point for subsequent refinement. This design enables efficient inference-time scaling, allowing generation quality to improve as the search budget increases, without additional training. Across multiple benchmarks, MEDAL achieves up to 22.0% improvement over existing inference strategies, establishing a new paradigm for search-based inference in DLMs.

Zhewen Xu, Baoxiang Pan, Hongliang Li et al.

The rapid development of large climate models has created the requirement of storing and transferring massive atmospheric data worldwide. Therefore, data compression is essential for meteorological research, but an efficient compression scheme capable of keeping high accuracy with high compressibility is still lacking. As an emerging technique, Implicit Neural Representation (INR) has recently acquired impressive momentum and demonstrates high promise for compressing diverse natural data. However, the INR-based compression encounters a bottleneck due to the sophisticated spatio-temporal properties and variability. To address this issue, we propose Hierarchical Harmonic decomposition implicit neural compression (HiHa) for atmospheric data. HiHa firstly segments the data into multi-frequency signals through decomposition of multiple complex harmonic, and then tackles each harmonic respectively with a frequency-based hierarchical compression module consisting of sparse storage, multi-scale INR and iterative decomposition sub-modules. We additionally design a temporal residual compression module to accelerate compression by utilizing temporal continuity. Experiments depict that HiHa outperforms both mainstream compressors and other INR-based methods in both compression fidelity and capabilities, and also demonstrate that using compressed data in existing data-driven models can achieve the same accuracy as raw data.