Graph-to-Sequence Generation on GenWiki (test)
0.748CrF++ ScoreDLM4G-2.o
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| DLM4G-2.o2026.04 | 0.748 | 0.469 | 0.574 | 0.899 | 0.892 | — | |
| DLM4G-2.oevaluation paradigm=Pretrain, #P=63M2026.04 | 0.748 | — | 0.574 | — | — | 0.469 | |
| DLM4G-2.oevaluation paradigm=Finetune, #P=63M2026.04 | 0.748 | — | 0.574 | — | — | 0.469 | |
| DLM4G-2.oevaluation paradigm=Zero-shot, #P=63M2026.04 | 0.748 | — | 0.574 | — | — | 0.469 | |
| GPT-o4-minievaluation paradigm=Zero-shot, #P=8B2026.04 | 0.734 | — | 0.471 | — | — | 0.464 | |
| LLaMa-3evaluation paradigm=Zero-shot, #P=8B2026.04 | 0.709 | — | 0.51 | — | — | 0.461 | |
| Qwen2.5evaluation paradigm=Zero-shot, #P=7B2026.04 | 0.697 | — | 0.501 | — | — | 0.461 | |
| DeepSeekevaluation paradigm=Zero-shot, #P=7B2026.04 | 0.688 | — | 0.533 | — | — | 0.391 | |
| DLM4G-1.o2026.04 | 0.663 | 0.401 | 0.527 | 0.857 | 0.841 | — | |
| DLM4G-1.oevaluation paradigm=Pretrain, #P=50M2026.04 | 0.663 | — | 0.527 | — | — | 0.401 | |
| Noisy-Sup.2026.04 | 0.623 | 0.384 | 0.414 | 0.878 | 0.901 | — | |
| T5 (L)evaluation paradigm=Finetune, #P=770M2026.04 | 0.567 | — | 0.338 | — | — | 0.361 | |
| T5 (S)evaluation paradigm=Finetune, #P=60M2026.04 | 0.495 | — | 0.447 | — | — | 0.227 | |
| Direct-Transfer2026.04 | 0.483 | 0.234 | 0.332 | 0.808 | 0.801 | — | |
| GPT-2 (B)evaluation paradigm=Finetune, #P=355M2026.04 | 0.47 | — | 0.425 | — | — | 0.312 | |
| GPT-2 (S)evaluation paradigm=Finetune, #P=124M2026.04 | 0.465 | — | 0.435 | — | — | 0.28 | |
| Rule-Based2026.04 | 0.36 | 0.219 | 0.397 | 0.679 | 0.822 | — |