lvchangze / snn Goto Github PK

Hello, this job is very enlightening to me. Meanwhile, I have paid attention to your latest work SpikeBERT, and I have some doubts.
In section 3.2 of this article you mentioned that the attention map in vanilla Spikformer is D × D where D is the dimensionality of the hidden layers. But I looked at the source code of the Spikformer, and did not find that its attention map is D × D, still N x N(N=H x W), I'd like to confirm this. Thanks!

Dear author, can you provide the content or format of the file sgns.merge.bigram in the main function of the data_preprocess/chinese_tensor_encoder.py file, a small demo is fine.

Dear authors, I was inspired and benefited from this article exploring the feasibility of using SNNs for natural language processing tasks. You guys are amazing! I don't know if I can ask a question, but when I reproduce the paper, I need a sgns.merge.bigram file, which is in data_preprocess/chinese_tensor_encoder.py file. I tried to generate a file with word2vec, but I can't run the program. Can you give me some format or content of the sgns.merge.bigram file, a small demo is fine.