Dialogue Generation on CausalDialogue (test)
1.2PPLHuman Written Responses
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Human Written Responses2022.12 | 1.2 | 48.9 | 34 | 25.9 | 1.7 | 11.1 | — | 100 | |
| T5Loss=MLE, Inference=Greedy Search2022.12 | 15.4 | 5.8 | 2.52 | 0.58 | 1.11 | 4.37 | 1.39 | 75.64 | |
| T5Loss=MLE, Inference=Softmax (T=0.5)2022.12 | 15.4 | 12.7 | 5.06 | 0.97 | 1.77 | 10.91 | 1.39 | 74.66 | |
| T5Loss=MLE, Inference=TopK (K=10)2022.12 | 15.4 | 14.1 | 5.09 | 0.82 | 2.07 | 15.49 | 1.39 | 72.79 | |
| T5Loss=ExMATE, Inference=Greedy Search2022.12 | 15.4 | 5.66 | 2.46 | 0.55 | 1.1 | 4.06 | 1.5 | 75.76 | |
| T5Loss=ExMATE, Inference=Softmax (T=0.5)2022.12 | 15.4 | 12.6 | 5.02 | 1 | 1.72 | 10.73 | 1.5 | 74.8 | |
| T5Loss=ExMATE, Inference=TopK (K=10)2022.12 | 15.4 | 14.1 | 5.06 | 0.8 | 2.06 | 15.67 | 1.5 | 72.83 | |
| DialoGPTLoss=MLE, Inference=Greedy Search2022.12 | 18.9 | 11.2 | 4.47 | 0.84 | 0.73 | 3.42 | 2.33 | 32.51 | |
| DialoGPTLoss=MLE, Inference=Softmax (T=0.5)2022.12 | 18.9 | 17 | 6.43 | 1.17 | 1.12 | 9.09 | 2.33 | 30.97 | |
| DialoGPTLoss=MLE, Inference=TopK (K=10)2022.12 | 18.9 | 15.7 | 5.34 | 0.81 | 1.37 | 13.57 | 2.33 | 27.65 | |
| DialoGPTLoss=ExMATE, Inference=Greedy Search2022.12 | 19 | 10.7 | 4.26 | 1.05 | 0.79 | 3.65 | 2.68 | 32.18 | |
| DialoGPTLoss=ExMATE, Inference=Softmax (T=0.5)2022.12 | 19 | 15.5 | 5.7 | 1.06 | 1.25 | 9.71 | 2.68 | 31.18 | |
| DialoGPTLoss=ExMATE, Inference=TopK (K=10)2022.12 | 19 | 13.5 | 4.47 | 0.67 | 1.52 | 14.44 | 2.68 | 28.16 |