Closed Woolseyyy closed 5 years ago
In the mxnet version and paper, it seems that embeddings are normalized before multiplied with normalized weight. Howerver, line 261 in model.py dosen't do so. Is this a bug?
I have found it normalized in the network.
In the mxnet version and paper, it seems that embeddings are normalized before multiplied with normalized weight. Howerver, line 261 in model.py dosen't do so. Is this a bug?