ChrisGeishauser's picture
Update README.md
9b9ce11
|
raw
history blame
704 Bytes
metadata
language:
  - en
license: apache-2.0
tags:
  - dialogue policy
  - task-oriented dialog
datasets:
  - ConvLab/multiwoz21

ddpt-policy-sgd

This is a DDPT model (https://aclanthology.org/2022.coling-1.21/) trained on MultiWOZ 2.1

Refer to ConvLab-3 for model description and usage.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 1e-05
  • train_batch_size: 64
  • seed: 1
  • optimizer: Adam
  • num_epochs: 40
  • use checkpoint which performed best on validation set

Framework versions

  • Transformers 4.18.0
  • Pytorch 1.10.2+cu111