Skip to content
This repository has been archived by the owner on Aug 28, 2021. It is now read-only.

Human chat with full model #27

Closed
rzhao1 opened this issue Feb 22, 2019 · 3 comments
Closed

Human chat with full model #27

rzhao1 opened this issue Feb 22, 2019 · 3 comments

Comments

@rzhao1
Copy link

rzhao1 commented Feb 22, 2019

Hi,

I have two questions:

  1. How can I make human chat with full model version. I noticed that chat.py is disabled.
  2. The dialogues generated by full model version is very short, even shorter than original RNN version. What is the reason?

Look forward to your reply!

@denisyarats
Copy link
Contributor

Hi,

  1. The chat.py does not support using the full model yet, but it should be straightforward to add.
  2. Perhaps the training parameters are not optimal. But in general, as stated in the paper (Table 5), the dialogues from the full model tend to be slightly shorter.

@rzhao1
Copy link
Author

rzhao1 commented Feb 25, 2019

Thank you fro your reply!

I remembered the "old negotiation model" normally generate 6.6 turns per dialogues. How about the "full model"?

@denisyarats
Copy link
Contributor

With RL or Rollouts it should be generating at least 5 turns on average.

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants