你好，我在复现你的代码时发现结果分为“l2r”和“r2l”两个方向，论文里只有一个方向，比如DBP15Kzh-en。请问论文里的结果是否是实验中“l2r”+“r2l”求平均得来的吗？

Hi，I don't konw how to combine "ref_ent_ids" and "sup_ent_ids" into a single file, could you give me a code example？I would appreciate it if you can help me.

Your work benefit me a lot! And i want to know the meaning of line 87 and 88 in layers.py.

if self.diag:
                h = torch.mul(input, self.w[i])

what is the diag means?Thanks a lot if you can give me explanations obout it.

Hi, thank you very much for sharing the code! It has been really useful!

The code runs perfectly on DBP15K. However, when I ran it on DWY100K, an error emerged, "CUDA out of memory. Tried to allocate 11.44 GiB (GPU 0; ...; 9.49 GiB free...)", which is expected, since the size of DWY100K is much larger than DBP15K, and one GPU might not be enough. Consequently, I tried to use torch.nn.DataParallel to run it on multiple GPUs, while it reports "Runtime Error : sparse tensors do not have strides". I wonder how you altered the code to make it work on DWY100K?

Thank you very much for your time and I look forward to your reply.

Thanks for uploading codes, I can't wait reading your paper.

I jumped to the link in README.md, but still can't find EMNLP paper.
Would you know where it is? If so, I would appreciate it much.

Thanks.