During March 2025, contributed to the keras-team/keras repository by developing a feature that enables cuDNN-accelerated training for recurrent neural networks when dropout is active. This work focused on updating the conditions under which cuDNN is used for GRU and LSTM layers, ensuring that dropout masks are correctly applied during training. By leveraging deep learning techniques and GPU acceleration with Python, the implementation allows for faster training throughput on GPUs without compromising model accuracy. The feature addressed a key limitation in existing RNN workflows, improving efficiency for users training recurrent models with dropout in Keras environments. No bugs were fixed.