Update README for 30min training"

Rewrite train.py: 30min max on free T4, 1000 samples, 1 epoch, G=2, 128 max tokens"

Add README with instructions"

Add GRPO training script for Colab/Kaggle free T4 GPU"

initial commit