wuch15 / fedkd Goto Github PK

1、运行run.py时，报错出现在 tuple index out of range:File "/data/liuqian/FedKD-main/model_bert.py", line 195, in forward word_emb = temp[3][self.nlayer] ；

2、使用的模型为bert-base-uncased，数据为MINDsmall_train;

3、因为temp的长度为3，最大索引为2，在计算 word_emb， word_emb2，word_att会报错tuple index out of range，是不是跟使用的模型有关，如果我使用的bert-base-uncased模型的话，这一块怎么取值？
print(len(temp))#len=3
word_emb = temp[3][self.nlayer]
word_emb2 = temp[3]
word_att=temp[4]

Thanks to the author for making the code open source. But there are some details about the code that I don't quite understand.

1. latest_local_student_para, which is not the latest parameter, seems to have been assigned at the beginning and not updated.

2. The aggregate parameter is student_update. The gradient is the difference between the latest weight parameter and the previous weight parameter. But when update a local model parameter, this code replace it directly, right? (Could the gradient be replaced directly with a weight value?)

I'm looking forward to your reply.

1. In run.py line 119 the code got the dataloader , however in dataloader line 161 , I find that all train data's label are set to 0. And in model_bert.py line 525, all targets are used to train model. So I want to know if it is a mistake?
2. MIND dataset is a recommendation task. However in run.py training process, I haven't find the use of behaviors.tsv but the only use of news.tsv. So I have the problem that what's the train target of the training process?

Thanks for such a wonderful code.
I have a probem about tnlrv3.
In code:
from tnlrv3.modeling import TuringNLRv3ForSequenceClassification
from tnlrv3.configuration_tnlrv3 import TuringNLRv3Config
from tnlrv3.tokenization_tnlrv3 import TuringNLRv3Tokenizer
In readme, I follow the descripion to find UniLM v2 in Microsoft, but I cannot find any codes includes the imported module above.
Could you please explain what exactly these three modules are imported or where should I go to find them？