Arithmetic Reasoning on Countdown 512 tokens
62.1Pass@1d-TreeRPO
Evaluation results
| Method | Links | |
|---|---|---|
| d-TreeRPOBase Model=LLaDA-8B-Instruct, Generation Length=5122025.12 | 62.1 | |
| d-TreeRPOBase Model=LLaDA-MoE-7BA1B-Instruct, Generation Length=5122025.12 | 60.6 | |
| GDPOBase Model=LLaDA-8B-Instruct, Generation Length=5122025.12 | 60.2 | |
| wd1Base Model=LLaDA-MoE-7BA1B-Instruct, Generation Length=5122025.12 | 58.7 | |
| SAPOBase Model=LLaDA-8B-Instruct, Generation Length=5122025.12 | 56.3 | |
| SAPOBase Model=LLaDA-MoE-7BA1B-Instruct, Generation Length=5122025.12 | 54.8 | |
| GDPOBase Model=LLaDA-MoE-7BA1B-Instruct, Generation Length=5122025.12 | 53.4 | |
| TraceRLBase Model=LLaDA-8B-Instruct, Generation Length=5122025.12 | 52.6 | |
| TraceRLBase Model=LLaDA-MoE-7BA1B-Instruct, Generation Length=5122025.12 | 49.1 | |
| Diffu-GRPOBase Model=LLaDA-MoE-7BA1B-Instruct, Generation Length=5122025.12 | 48.2 | |
| wd1Base Model=LLaDA-8B-Instruct, Generation Length=5122025.12 | 46.1 | |
| LLaDA-MoE-7BA1B-InstructBase Model=LLaDA-MoE-7BA1B-Instruct, Generation Length=5122025.12 | 41.4 | |
| Diffu-GRPOBase Model=LLaDA-8B-Instruct, Generation Length=5122025.12 | 37.1 | |
| VRPO (LLaDA-1.5)Base Model=LLaDA-8B-Instruct, Generation Length=5122025.12 | 18 | |
| LLaDA-8B-InstructBase Model=LLaDA-8B-Instruct, Generation Length=5122025.12 | 16 |