Qian Liu

Affiliations

Beihang UniversitySea AI Lab

papers

837

total citations

papers (10)

Agent Smith: A Single Image Can Jailbreak One Million Multimodal LLM Agents Exponentially Fast

ICML 2024arXiv

103

citations

Lemur: Harmonizing Natural Language and Code for Language Agents

ICLR 2024arXiv

citations

General-Reasoner: Advancing LLM Reasoning Across All Domains

NEURIPS 2025arXiv

citations

Compositional Generalization by Learning Analytical Expressions

NEURIPS 2020arXiv

citations

Longhorn: State Space Models are Amortized Online Learners

ICLR 2025arXiv

citations

Cheating Automatic LLM Benchmarks: Null Models Achieve High Win Rates

ICLR 2025arXiv

citations

SkyLadder: Better and Faster Pretraining via Context Window Scheduling

NEURIPS 2025arXiv

citations

Qian Liu

Affiliations

papers (10)

OctoPack: Instruction Tuning Code Large Language Models

Scaling up Masked Diffusion Models on Text

RegMix: Data Mixture as Regression for Language Model Pre-training

Agent Smith: A Single Image Can Jailbreak One Million Multimodal LLM Agents Exponentially Fast

Lemur: Harmonizing Natural Language and Code for Language Agents

General-Reasoner: Advancing LLM Reasoning Across All Domains

Compositional Generalization by Learning Analytical Expressions

Longhorn: State Space Models are Amortized Online Learners

Cheating Automatic LLM Benchmarks: Null Models Achieve High Win Rates

SkyLadder: Better and Faster Pretraining via Context Window Scheduling

papers (10)

OctoPack: Instruction Tuning Code Large Language Models

Scaling up Masked Diffusion Models on Text

RegMix: Data Mixture as Regression for Language Model Pre-training

Agent Smith: A Single Image Can Jailbreak One Million Multimodal LLM Agents Exponentially Fast

Lemur: Harmonizing Natural Language and Code for Language Agents

General-Reasoner: Advancing LLM Reasoning Across All Domains

Compositional Generalization by Learning Analytical Expressions

Longhorn: State Space Models are Amortized Online Learners

Cheating Automatic LLM Benchmarks: Null Models Achieve High Win Rates

SkyLadder: Better and Faster Pretraining via Context Window Scheduling