Problem of reproduce.

[shizizuoing]

Hello author, I have tried to reproduce this article many times, but have all received very low results. I have a few questions and hope to receive your results. I saw that the weight attenuation in your paper is 0.7, so I changed it to 0.7. No other hyperparameters have been modified. May I ask if this modification is correct? The result I obtained is similar to the image, which is different from the paper's result. This problem has puzzled me for a long time, and I hope to receive your help.

[Chen-Yang-Liu]

Hello author, I have tried to reproduce this article many times, but have all received very low results. I have a few questions and hope to receive your results. I saw that the weight attenuation in your paper is 0.7, so I changed it to 0.7. No other hyperparameters have been modified. May I ask if this modification is correct? The result I obtained is similar to the image, which is different from the paper's result. This problem has puzzled me for a long time, and I hope to receive your help.

Your modification is okay. You can try setting "--beam_size" to 3 and select the best model to run eval.py to get the metric. Besides, you can refer to this issue and try using our published model on your device to see if the score is normal.