liyucheng09 / contamination_detector

Lightweight tool to identify Data Contamination in LLMs evaluation

This is Yucheng

I am doing PhD on NLP, especially on Efficient LLMs and Robust LLMs evaluation, at University of Surrey, UK.

I am currently a Student Researcher at Microsoft Research.

Publicaitons list at: https://liyucheng09.github.io/

contamination_detector's Issues

Is the 2017 - 2020 contamination report available?

Hi,

Thanks for putting up this work and all these useful results. I'm particularly interested in LLaMA's contamination, and you mentioned that the corresponding search window for them was 2017 - 2020, different from other models' 2017 - 2023.

I was thus wondering if the reports under reports/ are for 2017 - 2020 or for 2017 - 2023? If it's the latter, are the reports for 2017 - 2020 available at the moment?

Thanks

Typos for Figure2(b)

This is a very interting work. But in Figure 2(b), there seems to be some confusion between the clean and memory icons for llama-7b, 13b, and 30b.