Question on Shared Weights Implementation?

Question

Question on Shared Weights Implementation?

truongthanhdat opened this issue 6 years ago · comments

The paper showed that weights of the last few layers of encoders and first few layers of decoders are shared. However, I have not found its implementations in the source code. It seems to be that two encoders (two decoders) use separate weights.

Maybe my limitation knowledge in Pytorch library makes me do not know where is the implementation of shared weights. Can you please indicate the implementation of shared weights constraint?

Thank you very much.

Ming-Yu Liu 劉洺堉 · Answer 1 · Wed Sep 05 2018 13:47:59 GMT+0800 (China Standard Time)

Please check the README.md for the code branch for getting the network used to produce the results in the NIPS paper.

Wang Dingpan · Answer 2 · Tue May 28 2019 17:56:04 GMT+0800 (China Standard Time)

Please check the README.md for the code branch for getting the network used to produce the results in the NIPS paper.

Hi, I have the same question about weight-sharing as the first questioner, I have also checked the README.md, but I still can't understand it. Could you explain it more detilely, please!