Running on Zero Agents Predicting Memorization Before Fine-Tuning 🚀 Predict memorization risk of text before fine‑tuning