Next step prediction on Gardening domain (test)
18.4BLEU-1BART+CP+M+R+CL
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| BART+CP+M+R+CLcontext-aware pretraining=true, Selective Multimedia Encoder=true, Retrieval-Augmented Decoder=true, Diversity-Oriented Contrastive Learning=true2022.08 | 18.4 | 9.72 | 5.51 | 3.31 | 9.91 | 17.3 | -4.37 | 0.31 | |
| BART+CP+Mcontext-aware pretraining=true, Selective Multimedia Encoder=true2022.08 | 17.8 | 9.36 | 5.3 | 3.19 | 9.61 | 17.4 | -4.38 | 0.305 | |
| T5number of parameters=222M2022.08 | 17.6 | 9.05 | 4.92 | 2.87 | 9.41 | 16.5 | -4.45 | 0.3 | |
| BART+CP+M+Rcontext-aware pretraining=true, Selective Multimedia Encoder=true, Retrieval-Augmented Decoder=true2022.08 | 17.5 | 9.22 | 5.25 | 3.13 | 9.6 | 17.2 | -4.36 | 0.309 | |
| BARTnumber of parameters=139M2022.08 | 17 | 8.21 | 4.45 | 2.61 | 8.93 | 15.7 | -4.52 | 0.277 | |
| BART+CPcontext-aware pretraining=true2022.08 | 16.9 | 8.79 | 4.99 | 3.03 | 9.23 | 16.5 | -4.41 | 0.3 | |
| Retrieval BARTbackbone=BART2022.08 | 16.8 | 8.68 | 4.8 | 2.24 | 9.15 | 16 | -4.43 | 0.295 | |
| CLIP-BARTbackbone=VL-BART, encoder=ViT-B/32 CLIP2022.08 | 14.4 | 7.1 | 3.77 | 2.22 | 8.28 | 13.8 | -4.44 | 0.256 | |
| GPT-22022.08 | 13.2 | 5.03 | 1.87 | 0.72 | 7.38 | 12.5 | -4.73 | 0.239 | |
| GPT2-SIFbackbone=GPT-22022.08 | 11.6 | 5.1 | 2.43 | 1.28 | 6.85 | 10.8 | -4.8 | 0.233 | |
| Naive Retrieval2022.08 | 10.9 | 4.14 | 1.93 | 1.1 | 6.33 | 10 | -4.88 | 0.18 |