Data-to-Text Generation on E2E (test)
68.23BLEUTransformer
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| TransformerImplementation=Authors' implementation2021.06 | 68.23 | 44.31 | — | 8.6765 | 69.88 | 2.2153 | — | — | — | — | — | |
| KGPT-Seqpre-training=true2020.10 | 68.05 | 45.8 | 70.92 | — | — | — | — | — | — | — | — | |
| KGPT-Graphpre-training=true2020.10 | 67.87 | 44.5 | 70 | — | — | — | — | — | — | — | — | |
| KGPT-Seqpre-training=false2020.10 | 67.67 | 45.33 | 70.39 | — | — | — | — | — | — | — | — | |
| Adapt2020.10 | 67.37 | 45.23 | 70.89 | — | — | — | — | — | — | — | — | |
| KGPT-Graphpre-training=false2020.10 | 66.47 | 44.2 | 67.78 | — | — | — | — | — | — | — | — | |
| TGenSource=Previous work2021.06 | 66.41 | 45.07 | — | 8.5565 | 69.17 | 2.2253 | — | — | — | — | — | |
| SLUG2SLUG2020.10 | 66.19 | 44.54 | 67.72 | — | — | — | — | — | — | — | — | |
| TGen2020.10 | 65.93 | 44.83 | 68.5 | — | — | — | — | — | — | — | — | |
| Shen et al.Year=2020, Source=Previous work2021.06 | 65.1 | 45.5 | — | — | 68.2 | 2.241 | — | — | — | — | — | |
| AGGGENInput Ordering=true, Input Aggregation=true2021.06 | 64.14 | 45.13 | — | 8.3509 | 66.62 | 2.1953 | — | — | — | — | — | |
| NTemp+ARSource=Previous work2021.06 | 59.8 | 38.75 | — | 7.56 | 65.01 | 1.95 | — | — | — | — | — | |
| AGGGEN-ODInput Ordering=shuffled, Input Aggregation=true2021.06 | 58.9 | 43.21 | — | 7.91 | 62.12 | 1.9656 | — | — | — | — | — | |
| NTemp2020.10 | 55.17 | 38.75 | 65.01 | — | — | — | — | — | — | — | — | |
| C-P (large)Linear=O, Training Protocol=Multi-task2023.08 | 44.2 | — | — | — | — | — | — | — | — | — | — | |
| AGGGEN-AGInput Ordering=true, Input Aggregation=false2021.06 | 44 | 43.75 | — | 6.089 | 58.24 | 0.8202 | — | — | — | — | — | |
| T5-3BLinear=Uni, Training Protocol=Multi-task2023.08 | 43.2 | — | — | — | — | 1.99 | — | — | — | — | — | |
| T5-baseLinear=Uni, Training Protocol=Multi-task2023.08 | 42.9 | — | — | — | — | 1.94 | — | — | — | — | — | |
| T5-3BLinear=Uni, Training Protocol=Single-task2023.08 | 42.8 | — | — | — | — | 1.92 | — | — | — | — | — | |
| T5-3BLinear=O, Training Protocol=Single-task2023.08 | 42.5 | — | — | — | — | 1.94 | — | — | — | — | — | |
| T5-baseLinear=O, Training Protocol=Single-task2023.08 | 42.1 | — | — | — | — | 1.91 | — | — | — | — | — | |
| T5-baseLinear=Uni, Training Protocol=Single-task2023.08 | 41.8 | — | — | — | — | 1.9 | — | — | — | — | — | |
| T5-3BLinear=O, Training Protocol=Multi-task2023.08 | 41.8 | — | — | — | — | 1.89 | — | — | — | — | — | |
| T5-baseLinear=O, Training Protocol=Multi-task2023.08 | 41.7 | — | — | — | — | 1.89 | — | — | — | — | — | |
| SVD DenoisingModel=Qwen, Size=4B2025.10 | 40.88 | 66.3 | — | 5.03 | 55.74 | 1.36 | — | — | — | — | — | |
| DP-LoRAModel=Qwen, Size=4B2025.10 | 40 | 65.9 | — | 4.83 | 55.19 | 1.27 | — | — | — | — | — | |
| SVD DenoisingModel=Qwen, Size=0.6B2025.10 | 37.35 | 62.8 | — | 4.6 | 53.52 | 1.11 | — | — | — | — | — | |
| SVD DenoisingModel=Qwen, Size=1.7B2025.10 | 36.95 | 62.4 | — | 4.71 | 54.42 | 1.14 | — | — | — | — | — | |
| SVD DenoisingModel=Llama, Size=1B2025.10 | 36.33 | 62.6 | — | 4.62 | 52.86 | 1.13 | — | — | — | — | — | |
| DP-LoRAModel=Qwen, Size=1.7B2025.10 | 36.01 | 60.9 | — | 4.64 | 53.57 | 1.08 | — | — | — | — | — | |
| OracleModel=BART-base, Data cleaning=Handcrafted rules2022.12 | 35.42 | — | — | — | 54.44 | — | 1.43 | — | — | — | — | |
| CEAModel=BART-base, Data cleaning=Contrastive Error Attribution2022.12 | 35.19 | — | — | — | 54.19 | — | 2.76 | — | — | — | — | |
| TracInModel=BART-base, Data cleaning=Influence-based attribution2022.12 | 34.9 | — | — | — | 54.1 | — | 5.08 | — | — | — | — | |
| BaselineModel=BART-base, Data cleaning=None (trained on entire training set)2022.12 | 33.81 | — | — | — | 53.42 | — | 6.08 | — | — | — | — | |
| Fully-supervised fine-tuningTraining setting=Fully-supervised2023.05 | 29.35 | 52.9 | — | — | 50.93 | — | — | 69.77 | 42.87 | 94.76 | 41.91 | |
| CycleNLGTraining setting=Low-resource cycle training2023.05 | 29.22 | 53.02 | — | — | 50.51 | — | — | 69.53 | 42.48 | 94.74 | 41.39 | |
| LAP2Noise Type=LAP2, epsilon=1.2, Backbone=DistilGPT-2, Batch size=80, Clipping value=22026.02 | 27.99 | 20.5 | — | 4.4001 | 36.85 | 0.5301 | — | — | — | — | — | |
| CycleNLGTraining setting=Unsupervised cycle training2023.05 | 27.92 | 50.49 | — | — | 45.96 | — | — | 63.43 | 37.73 | 93.71 | 37.97 | |
| LAP2Noise Type=LAP2, epsilon=1, Backbone=DistilGPT-2, Batch size=80, Clipping value=22026.02 | 27.34 | 20.2 | — | 4.31 | 36.62 | 0.5152 | — | — | — | — | — | |
| LAP2Noise Type=LAP2, epsilon=0.4, Backbone=DistilGPT-2, Batch size=80, Clipping value=22026.02 | 26.9 | 19.8 | — | 4.12 | 34.2 | 0.465 | — | — | — | — | — | |
| SVD DenoisingModel=Llama, Size=3B2025.10 | 26.33 | 51.6 | — | 3.48 | 44.36 | 0.64 | — | — | — | — | — | |
| Low-resource fine-tuning + additional pre-trainingTraining setting=Low-resource, Additional pre-training=true2023.05 | 26.29 | 50.11 | — | — | 48.65 | — | — | 66.88 | 39.45 | 94.35 | 39.65 | |
| Low-resource fine-tuningTraining setting=Low-resource2023.05 | 25.31 | 48.8 | — | — | 48.59 | — | — | 66.62 | 39.68 | 94.35 | 39.56 | |
| GaussianNoise Type=Gaussian, epsilon=1.2, Backbone=DistilGPT-2, Batch size=80, Clipping value=22026.02 | 23.39 | 16.84 | — | 3.9474 | 33.49 | 0.3385 | — | — | — | — | — | |
| DP-LoRAModel=Qwen, Size=0.6B2025.10 | 23.17 | 50.9 | — | 2.34 | 46.78 | 0.67 | — | — | — | — | — | |
| GaussianNoise Type=Gaussian, epsilon=1, Backbone=DistilGPT-2, Batch size=80, Clipping value=22026.02 | 22.83 | 16.73 | — | 3.5342 | 32.88 | 0.3232 | — | — | — | — | — | |
| DP-LoRAModel=Llama, Size=1B2025.10 | 22.15 | 47.6 | — | 2.59 | 44.24 | 0.47 | — | — | — | — | — | |
| GaussianNoise Type=Gaussian, epsilon=0.4, Backbone=DistilGPT-2, Batch size=80, Clipping value=22026.02 | 16.16 | 15.2 | — | 2.7025 | 32.5 | 0.2586 | — | — | — | — | — | |
| DP-LoRAModel=Llama, Size=3B2025.10 | 10.96 | 26.8 | — | 0.93 | 25.22 | 0.22 | — | — | — | — | — | |
| BEST-ON-VALEnsemble Size=3B2024.12 | — | — | — | — | — | — | — | — | 31.1 | — | — | |
| BEST-ON-VALEnsemble Size=7B2024.12 | — | — | — | — | — | — | — | — | 36.7 | — | — | |
| RANDOMEnsemble Size=3B2024.12 | — | — | — | — | — | — | — | — | 27.3 | — | — | |
| RANDOMEnsemble Size=7B2024.12 | — | — | — | — | — | — | — | — | 35.3 | — | — | |
| SMOOTHIE-GLOBALEnsemble Size=3B2024.12 | — | — | — | — | — | — | — | — | 31.8 | — | — | |
| SMOOTHIE-GLOBALEnsemble Size=7B2024.12 | — | — | — | — | — | — | — | — | 36.9 | — | — |