您好！看了你的论文 ReCoSa: Detecting the Relevant Contexts with Self-Attention for
Multi-turn Dialogue Generation，就直接过来了想看看您的源码，不知道什么时候可以release？

首先感谢作者公开的源码
源代码中是使用tensorflow写的，但是我对tensorflow并不熟悉
请问作者 我是否可以 使用pytorch的transformer 在最前面加一层lstm encoder层
万分感谢

I can not find the code for this part. Do you give me some informaiton about this? Many thanks.

We conduct the Ubuntu experiment follow your pipeline, but we get "ValueError: Cannot create a tensor proto whose content is larger than 2GB". We found that you convert the whole training data to tensor in load_data.py:

def get_batch_data():

# Load data
X,X_length, Y, sources,targets = load_train_data()
# calc total batch count
num_batch = len(X) // hp.batch_size
# Convert to tensor
X = tf.convert_to_tensor(X, tf.int32)
Y = tf.convert_to_tensor(Y, tf.int32)

We are sure that the Ubuntu dataset is mush larger than 2GB, so we are confused how did you do the Ubuntu experiment?

论文中提到的数据集，在代码中没有看到。麻烦问下数据集可以公开吗？

你好，请问可以看一下数据集文件的格式吗？只几行就可以，中英文的都可以，最好的中文的数据集格式，谢谢！

您好，请问论文中用到的京东对话数据集有什么方式获取呢，目前京东官方已经不提供下载了

你好！十分感谢你的开源代码。只是在阅读时，有一处细节不太了解——在模型输入时，你将多轮的上下文拆解成多个样本。即，

The dialogue data：Hello How are you? Good, you? I'm fine, what's new?
Souce looks like:

Hello How are you?

Hello How are you? Good, you?

Hello How are you? Good, you? I'm fine, what's new?

Target:

Good, you?

I'm fine, what's new?

Nothing much...

请问，这是多轮对话的通用处理方式吗？还是直接将多轮的上下文作为输入，也可以？

可能这个问题略显幼稚，但还是期待你的回复。

Hi, thanks for your open source codes of this work.
I try to apply your code on a new dataset DialogDialog, but I found that the outputs of the model are all the token '.' which means nothing.

So, I'm very curious that if this code is not appropriate to other datasets?
Can you help me troubleshoot the issue?

I have trained 200 epochs with ReCoSa model and the result as the following:

could this be a data format problem?

Hi. When I employ the Transformer Encoder-Decoder framework on dialogue generation task, there are two serious issues on the generated results. First, the generated results are always begin with word 'i'; second, the repetition badly weakens the performance.

Target: hmm i guess on the 28th
Predict: i i am i the bus of and and and and and and are and and and and and and and and and and and

The above is a typical bad example. Did you face such issue? Could you provide some suggestions to handle this problem? Thank you.