1 paper · 1 filter
Muhammed Yusuf Kocyigit, Caglar Yildirim
We present a controlled study of how dataset contamination interacts with the post-training stages now standard in large language model training pipelines. Starting from clean chec…