We implemented the training process of the KTAE algorithm by rewriting the code of the verl training framework. However, due to anonymity requirements, we only provided the core code of the KTAE algorithm. We will open source the rewritten complete training framework later. In addition, we also provided the results saved in the 'eval_result' file when doing comparative experiments with the baseline method. The 'eval' folder contains the code and benchmark we use to evaluate the model.