awslabs / pptod

Multi-Task Pre-Training for Plug-and-Play Task-Oriented Dialogue System (ACL 2022)
https://arxiv.org/abs/2109.14739
Apache License 2.0
158 stars 27 forks source link

About E2E_Tod Modeling #23

Closed Salierioo closed 1 year ago

Salierioo commented 1 year ago

When perform inference using pretrained checkpoints, I found the NLG in function e2e_batch_generate using the encoded golden context as inputs. In other words, the model infers with the golden previous system responses instead of the system responses generated by model.

I don't know if this problem is due to a misunderstanding that I didn't read the code carefully.