We are already building the encoder-decoder model and training the data on it. Then we are saving the trained model weights for prediction purpose.
Then why are we again creating the inference model for the prediction time? We should use the trained model or its trained weights in the inference time, right?
Hi
Your paper"English-Marathi Neural Machine Translation Using Local Attention" looks very interesting. Could you please tell me if this code implements local attention for machine translation?
Thank you in advance