Xipeng Qiu

Full Professor@Shanghai Innovation Institution·中国·OpenReview

研究方向

Large Multi-modal Model · Large Language Models · Information Extraction · Named Entity Recognition · Language Modeling · Pre-trained Models · Deep Learning · Natural Language Processing

Xipeng Qiu

FutureX: An Advanced Live Benchmark for LLM Agents in Future Prediction

FASTer: Toward Powerful and Efficient Autoregressive Vision–Language–Action Models with Learnable Action Tokenizer and Block-wise Decoding

Towards Understanding the Nature of Attention with Low-Rank Sparse Decomposition

Beyond Real: Imaginary Extension of Rotary Position Embeddings for Long-Context LLMs

RoboOmni: Proactive Robot Manipulation in Omni-modal Context

YuE: Scaling Open Foundation Models for Long-Form Music Generation

SIM-CoT: Supervised Implicit Chain-of-Thought

Towards True Speech-to-Speech Models Without Text Guidance

Game-RL: Synthesizing Multimodal Verifiable Game Data to Boost VLMs' General Reasoning

Evolution of Concepts in Language Model Pre-Training

Explicit Multi-head Attention for Inter-head Interaction in Large Language Models

Teach2Eval: An Interaction-Driven LLMs Evaluation Method via Teaching Effectiveness

Beyond Homogeneous Attention: Memory-Efficient LLMs via Fourier-Approximated KV Cache

A Benchmark for Self-Evolving Agents via Experience-Driven Lifelong Learning

CaTok: Taming Mean Flows for One-Dimensional Causal Image Tokenization

Attention Layers Add Into Low-Dimensional Residual Subspaces

Search or Think? Rethinking Iterative RAG from An Entropy Perspective

XY-Tokenizer: Mitigating the Semantic-Acoustic Conflict in Low-Bitrate Speech Codecs

Sparser Block-Sparse Attention via Token Permutation

In-depth Robustness Analysis for Vision-Language-Action Models

Inference-Time Alignment Control for Diffusion Models with Reinforcement Learning Guidance

Model Utility Law: Evaluating LLMs beyond Performance via Mechanistically Interpretable Metric

Speech-CLAP: Towards Style-Aware Speech Representation

VideoRoPE: What Makes for Good Video Rotary Position Embedding?

Pre-Trained Policy Discriminators are General Reward Models

Data Mixing Laws: Optimizing Data Mixtures by Predicting Language Modeling Performance

BitStack: Any-Size Compression of Large Language Models in Variable Memory Environments

Towards Universality: Studying Mechanistic Similarity Across Language Model Architectures

ForgerySleuth: Empowering Multimodal Large Language Models for Image Manipulation Detection

World-aware Planning Narratives Enhance Large Vision-Language Model Planner

Implicit Reward as the Bridge: A Unified View of SFT and DPO Connections

CHiP: Cross-modal Hierarchical Direct Preference Optimization for Multimodal LLMs

Domain2Vec: Vectorizing Datasets to Find the Optimal Data Mixture without Training

ReAttention: Training-Free Infinite Context with Finite Attention Scope

AgentGym: Evaluating and Evolving Large Language Model-based Agents across Diverse Envronments

INST-IT: Boosting Instance Understanding via Explicit Visual Prompt Instruction Tuning

Unearthing Large Scale Domain-Specific Knowledge from Public Corpora

Understanding the Role of LLMs in Multimodal Evaluation Benchmarks

Domain2Vec: Vectorizing Datasets to Find the Optimal Data Mixture without Training

LongSafetyBench: Long-Context LLMs Struggle with Safety Issues

Automatically Identifying and Interpreting Sparse Circuits with Hierarchical Tracing