VectorGym: A Multitask Benchmark for SVG Code Generation, Sketching, and Editing
为解决SVG生成、编辑等缺乏专业级基准问题，提出VectorGym，采用多任务强化学习方法优化，并提供人类标注数据和评估指标。
3 citationsRead paper
Selected work
Representative Papers
Fixed State, Long Reach: What a Constant-Size Cache Buys Block Diffusion at Scale
研究通过使用固定大小的缓存解决了大规模块扩散模型中的内存和延迟问题，采用序列混合器和块因果训练目标，显著降低了内存需求和提高了处理速度。
0 citationsRead paper
DARTS: Decoder-Aware Representation Tuning via Surgery for Model Merging
本文针对解码器模型在合并时出现的表示偏差问题，提出DARTS方法，通过引入熵加权L1损失和位置依赖偏置来校正，从而改善模型性能。
0 citationsRead paper
AgentJudgeBench: A Multi-Difficulty Benchmark for Evaluating LLM Judges on Agentic Tool-Calling
研究通过AgentJudgeBench评估LLM法官在处理依赖驱动工作流时的可靠性问题，揭示了任务难度对模型表现的影响及提高一致性的方法。
0 citationsRead paper
StarHarness: Evolving Harnesses with Stratified Search for Enterprise Environments
StarHarness通过分层搜索进化环境特定的代理框架，解决模型与企业环境不匹配问题，提高基准性能20-35个百分点。
0 citationsRead paper
Recent publications
Latest Papers
Fixed State, Long Reach: What a Constant-Size Cache Buys Block Diffusion at Scale
研究通过使用固定大小的缓存解决了大规模块扩散模型中的内存和延迟问题，采用序列混合器和块因果训练目标，显著降低了内存需求和提高了处理速度。
0 citationsRead paper
DARTS: Decoder-Aware Representation Tuning via Surgery for Model Merging
本文针对解码器模型在合并时出现的表示偏差问题，提出DARTS方法，通过引入熵加权L1损失和位置依赖偏置来校正，从而改善模型性能。
0 citationsRead paper
AgentJudgeBench: A Multi-Difficulty Benchmark for Evaluating LLM Judges on Agentic Tool-Calling
研究通过AgentJudgeBench评估LLM法官在处理依赖驱动工作流时的可靠性问题，揭示了任务难度对模型表现的影响及提高一致性的方法。
0 citationsRead paper
StarHarness: Evolving Harnesses with Stratified Search for Enterprise Environments
StarHarness通过分层搜索进化环境特定的代理框架，解决模型与企业环境不匹配问题，提高基准性能20-35个百分点。
0 citationsRead paper
Backdoor Decontamination Dynamics in LLM Agents
This study addresses the vulnerability of open-source large language model agents to stealthy backdoors implanted during fine-tuning, which are difficult to detect when trigger conditions remain unobserved. The work systematically investigates the efficacy of defensive poisoning and unlearning in mitigating unknown backdoors, revealing for the first time that trigger recognition and malicious execution can be behaviorally decoupled. It proposes a novel strategy: applying defensive poisoning with analogous triggers followed by depoisoning, which nearly eliminates the original backdoor. Evaluated on the AgentDyn framework with J-lens representation visualization across 115 experiments, defensive poisoning alone removes approximately 56% of backdoors, while combining it with depoisoning achieves near-complete (≈100%) removal. Notably, in multi-backdoor settings, neutralizing one known backdoor incidentally eradicates 87% of coexisting unknown backdoors.
0 citationsRead paper