22 papers
QRAE: Query-Relevant Attention Entropy for Context Compression in DeepSearch Agents
Yingjie Tan, Haiquan Qiu, Quanming Yao +3
DeepSearch LLM agents repeatedly search and carry retrieved sources into subsequent reasoning steps. This expanding context increases token consumption and latency, although many r…
Experience Funnel: A State-Policy Alternating Loop for Self-Evolving Agents
Wenbo Gao, Zhaomou Song, Zhiyuan Ji +7
Autonomous agents powered by large language models (LLMs) continuously accumulate experience through interaction, creating an opportunity to improve future behavior through self-ev…
How Does mHC Use Its Residual Streams? Selective Routing and Near-Identity Mixing
Pengxiang Zhao, Xing Li, Xianzhi Yu +2
Hyper-Connections and their manifold-constrained variant mHC widen a residual pathway from one stream to n, yet how trained models use this capacity remains unclear: how broadly bl…
LEGO-RL: Harness-Native Reinforcement Learning for Coding Agents
Yiming Du, Yuxin Jiang, Tao Yuan +9
Reinforcement learning for coding agents increasingly relies on long-running agent harnesses to manage tool integration, repository contexts, and execution feedback. However, the n…
PSD: Pushing the Pareto Frontier of Diffusion LLMs via Parallel Speculative Decoding
Shengyin Sun, Yiming Li, Renxi Liu +7
Diffusion large language models (dLLMs) generate text by iteratively denoising masked token sequences. Although dLLMs can predict all masked positions in parallel within each step,…
FocuSFT: Bilevel Optimization for Dilution-Aware Long-Context Fine-Tuning
Zehua Pei, Hui-Ling Zhen, Xianzhi Yu +3
Large language models can now process increasingly long inputs, yet their ability to effectively use information spread across long contexts remains limited. We trace this gap to h…
DLLM Agent: See Farther, Run Faster
Huiling Zhen, Weizhe Lin, Renxi Liu +15
Diffusion large language models (DLLMs) have emerged as an alternative to autoregressive (AR) decoding with appealing efficiency and modeling properties, yet their implications for…
Unleashing Low-Bit Inference on Ascend NPUs: A Comprehensive Evaluation of HiFloat Formats
Pengxiang Zhao, Hui-Ling Zhen, Xing Li +10
As LLMs scale, low-bit floating-point formats like MXFP and NVFP4 offer new opportunities for precision and efficiency. In this work, we evaluate HiFloat (HiF8 and HiF4), a family…
What Matters For Safety Alignment?
Xing Li, Hui-Ling Zhen, Lihao Yin +3
This paper presents a comprehensive empirical study on the safety alignment capabilities. We evaluate what matters for safety alignment in LLMs and LRMs to provide essential insigh…
JuDi: Revisiting Judge Decoding from First Principles via Training-Free Distributional Divergence
Shengyin Sun, Yiming Li, Renxi Liu +5
Judge Decoding accelerates LLM inference by relaxing the strict verification of Speculative Decoding, yet it typically relies on expensive and noisy supervision. In this work, we r…
SwiftMem: Fast Agentic Memory via Query-aware Indexing
Anxin Tian, Yiming Li, Xing Li +5
Agentic memory systems have become critical for enabling LLM agents to maintain long-term context and retrieve relevant information efficiently. However, existing memory frameworks…
Towards Efficient Agents: A Co-Design of Inference Architecture and System
Weizhe Lin, Hui-Ling Zhen, Shuai Yang +14
The rapid development of large language model (LLM)-based agents has unlocked new possibilities for autonomous multi-turn reasoning and tool-augmented decision-making. However, the…
Behavioral Fingerprinting of Large Language Models
Zehua Pei, Hui-Ling Zhen, Ying Zhang +5
Current benchmarks for Large Language Models (LLMs) primarily focus on performance metrics, often failing to capture the nuanced behavioral characteristics that differentiate them.…
Scaling Up, Speeding Up: A Benchmark of Speculative Decoding for Efficient LLM Test-Time Scaling
Shengyin Sun, Yiming Li, Xing Li +8
Test-time scaling has emerged as a powerful paradigm for enhancing the reasoning capabilities of large language models (LLMs) by allocating additional computational resources durin…
HAP: Hybrid Adaptive Parallelism for Efficient Mixture-of-Experts Inference
Haoran Lin, Xianzhi Yu, Kang Zhao +7
Current inference systems for Mixture-of-Experts (MoE) models primarily employ static parallelization strategies. However, these static approaches cannot consistently achieve optim…
EAQuant: Enhancing Post-Training Quantization for MoE Models via Expert-Aware Optimization
Zhongqian Fu, Tianyi Zhao, Ning Ding +4
Mixture-of-Experts (MoE) models enable scalable computation and performance in large-scale deep learning but face quantization challenges due to sparse expert activation and dynami…
TrimR: Verifier-based Training-Free Thinking Compression for Efficient Test-Time Scaling
Weizhe Lin, Xing Li, Zhiyuan Yang +7
Large Reasoning Models (LRMs) demonstrate exceptional capability in tackling complex mathematical, logical, and coding tasks by leveraging extended Chain-of-Thought (CoT) reasoning…
PreMoE: Proactive Inference for Efficient Mixture-of-Experts
Zehua Pei, Ying Zhang, Hui-Ling Zhen +6
Mixture-of-Experts (MoE) models offer dynamic computation, but are typically deployed as static full-capacity models, missing opportunities for deployment-specific specialization.…
SVDq: 1.25-bit and 410x Key Cache Compression for LLM Attention
Hong Yankun, Li Xing, Zhen Hui-Ling +3
For the efficient inference of Large Language Models (LLMs), the effective compression of key-value (KV) cache is essential. Three main types of KV cache compression techniques, na…
Analytical FFN-to-MoE Restructuring via Activation Pattern Analysis
Zehua Pei, Hui-Ling Zhen, Lancheng Zou +5
Scaling large language models (LLMs) improves performance but significantly increases inference costs, with feed-forward networks (FFNs) consuming the majority of computational res…
AttentionPredictor: Temporal Patterns Matter for KV Cache Compression
Qingyue Yang, Jie Wang, Xing Li +8
With the development of large language models (LLMs), efficient inference through Key-Value (KV) cache compression has attracted considerable attention, especially for long-context…
From Pruning to Grafting: Dynamic Knowledge Redistribution via Learnable Layer Fusion
Zehua Pei, Hui-Ling Zhen, Xianzhi Yu +3
Structured pruning of Generative Pre-trained Transformers (GPTs) offers a promising path to efficiency but often suffers from irreversible performance degradation due to the discar…