Unconditional Language Generation on ROCStories (test)
0.951MAUVEReference
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Reference2022.12 | 0.951 | 21.1 | 0.414 | 0.362 | |
| GPT-2 MediumTraining=Fine-tuned, Sampling=Nucleus (p=0.95)2022.12 | 0.788 | 20 | 0.372 | 0.688 | |
| Diffusion-LMSampling method=DDIM-style2026.01 | 0.76 | 21.05 | 0.18 | 0.14 | |
| LD4LGBackbone=BART-Base, Timesteps=2502022.12 | 0.716 | 30.6 | 0.331 | 0.441 | |
| MuLAN-RescaledSampling method=DDIM-style2026.01 | 0.6 | 27.71 | 0.2 | 0.12 | |
| LD4LGBackbone=FLAN-T5-base, Timesteps=2502022.12 | 0.481 | 37.5 | 0.389 | 0.387 | |
| NFDMSampling method=DDIM-style2026.01 | 0.43 | 26.44 | 0.11 | 0.14 | |
| Diffusion-LMTimesteps=20002022.12 | 0.043 | 47.3 | 0.128 | 0.434 | |
| MuLANSampling method=DDIM-style2026.01 | 0.01 | 139.74 | 0.21 | 0.02 |