You could visualize contamination examples after generating the reports via:

python visualize/visualize_search.py

This will highlight the matched part of benchmark samples.

MMLU - An example of contamination visualization

Question: The economy is in a deep recession. Given this economic situation which of the following statements about monetary policy is accurate?

Matches:

It will hightlight the overlapping part of the benchmark and internet pages.

Check more contamination examples: MMLU at here, and C-Eval at here

You could generate the figure to visualize the perplexity results via:

python visualize/visualize_perplexity.py

Check perplexity analysis example of QA (BoolQ, SQuAD, QuAD) benchmarks here