ResearchBenchmarksGoal-oriented dialogue on WOZ (test)Follow84.9dsEMGPT-284.3884.51584.6584.785Sep 7, 2019Evaluation ResultsMethodMethodLinksdsEMGPT-2Backbone=Smallest pre-...Backbone=Smallest pre-trained GPT-2 model, Training epochs=9, Decoding=Greedy2019.0984.9MQANEvaluation mode=Single...Evaluation mode=Single task scores2019.0984.4