End-to-end tool-graph planning on TaskBench-23 2024 (N=501)
73.4ToolF1LLaDA proposer
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LLaDA proposerProposer=LLaDA proposer, Refiner=AR refiner2026.06 | 73.4 | 25.8 | |
| Dream proposerProposer=Dream proposer, Refiner=AR refiner2026.06 | 72.9 | 26.8 | |
| AR proposerProposer=AR proposer, Refiner=AR refiner2026.06 | 66.1 | 24.2 | |
| Retriever (top-k=15)Proposer=Retriever, Refiner=AR refiner, k=152026.06 | 66 | 18.5 | |
| Retriever (top-k=10)Proposer=Retriever, Refiner=AR refiner, k=102026.06 | 63.8 | 15.6 | |
| AR-onlyProposer=AR-only, Refiner=None2026.06 | 62.9 | 20 | |
| Retriever (top-k=5)Proposer=Retriever, Refiner=AR refiner, k=52026.06 | 56.9 | 12.7 | |
| Dream DLM-onlyProposer=Dream DLM-only, Refiner=None2026.06 | 12.8 | 33.5 |