End-to-End Task-Oriented Dialog on In-Car
90.58Match RateESAinsTOD
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| ESAinsTODBackbone=Llama 2 7B2026.03 | 90.58 | 88.09 | 27.87 | 117.21 | |
| LABESpre-training=None2021.11 | 85.8 | 77 | 22.8 | 104.2 | |
| LABES2026.03 | 85.8 | 77 | 22.8 | 104.2 | |
| GALAXYPre-training initialization=Proposed dialogue pre-training2021.11 | 85.3 | 83.6 | 23 | 107.45 | |
| SPACEFine-tuned individually=true2026.03 | 85.26 | 83.16 | 22.92 | 107.13 | |
| FSDMpre-training=None2021.11 | 84.8 | 82.1 | 21.5 | 104.95 | |
| FSDM2026.03 | 84.8 | 82.1 | 21.5 | 104.95 | |
| SEDSTpre-training=None2021.11 | 84.5 | 82.9 | 19.3 | 103 | |
| TSCPpre-training=None2021.11 | 84.5 | 81.1 | 21.9 | 104.7 | |
| SEDST2026.03 | 84.5 | 82.9 | 19.3 | 103 | |
| TSCP2026.03 | 84.5 | 81.1 | 21.9 | 104.7 | |
| GALAXYPre-training initialization=original weights of UniLM2021.11 | 81.9 | 83.3 | 22 | 104.6 |