GALAXY: A Generative Pre-trained Model for Task-Oriented Dialog with Semi-Supervised Learning and Explicit Policy Injection

Open paper
Benchmark Model Rank Results
end-to-end-dialogue-modelling-on-multiwoz-2-0GALAXY#1MultiWOZ (Success): 85.3MultiWOZ (Inform): 94.4BLEU: 20.5
end-to-end-dialogue-modelling-on-multiwoz-2-1GALAXY#1BLEU: 20.01MultiWOZ (Inform): 95.30MultiWOZ (Success): 86.20

Rank counts only results with a code link.