When Benchmarks Leak: Inference-Time Decontamination for LLMs

Documenti analoghi

• One Leak Away: How Pretrained Model Exposure Amplifies Jailbreak Risks in Finetuned LLMs
  di: Tan, Yixin, et al.
  Pubblicazione: (2025)
• Inference-Time Decontamination: Reusing Leaked Benchmarks for Large Language Model Evaluation
  di: Zhu, Qin, et al.
  Pubblicazione: (2024)
• Leak, Cheat, Repeat: Data Contamination and Evaluation Malpractices in Closed-Source LLMs
  di: Balloccu, Simone, et al.
  Pubblicazione: (2024)
• When LLMs Benchmark Themselves: Deconstructing Self-Bias in Automated Evaluation
  di: Xu, Wenda, et al.
  Pubblicazione: (2025)
• LessLeak-Bench: A First Investigation of Data Leakage in LLMs Across 83 Software Engineering Benchmarks
  di: Zhou, Xin, et al.
  Pubblicazione: (2025)