ddpt-policy-sgd_0.01multiwoz21

This is a DDPT model (https://aclanthology.org/2022.coling-1.21/) trained on Schema-Guided Dialog and afterwards on 1 percent of 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: 0
• optimizer: Adam
• num_epochs: 40
• use checkpoint which performed best on validation set

Framework versions

• Transformers 4.18.0
• Pytorch 1.10.2+cu111