Text Generation on OpenWebText (OWT) GPT-2 tokenizer (val)
15.36PPLMDLM-Prime
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MDLM-Primel=6, Autoregressive formulation=false, Architecture=DiT-B, Parameters=130M2025.05 | 15.36 | 0.25 | |
| MDLM-Primel=8, Autoregressive formulation=false, Architecture=DiT-B, Parameters=130M2025.05 | 15.48 | 0.03 | |
| MDLM-Primel=4, Autoregressive formulation=false, Architecture=DiT-B, Parameters=130M2025.05 | 15.62 | 1.83 | |
| MDLM-Primel=3, Autoregressive formulation=false, Architecture=DiT-B, Parameters=130M2025.05 | 16.36 | 4.97 | |
| ARMAutoregressive formulation=true, Architecture=DiT-B, Parameters=130M2025.05 | 17.54 | — | |
| EDLM-coARAutoregressive formulation=true, Architecture=DiT-B, Parameters=130M2025.05 | 17.58 | — | |
| MDLM-Primel=2, Autoregressive formulation=false, Architecture=DiT-B, Parameters=130M2025.05 | 17.9 | 13.52 | |
| BD3-LMsAutoregressive formulation=true, Architecture=DiT-B, Parameters=130M2025.05 | 20.73 | — | |
| EDLM-NCEAutoregressive formulation=false, Architecture=DiT-B, Parameters=130M2025.05 | 21.52 | — | |
| GenMD4Autoregressive formulation=false, Architecture=DiT-B, Parameters=130M2025.05 | 21.8 | — | |
| MDLMAutoregressive formulation=false, Architecture=DiT-B, Parameters=130M2025.05 | 22.98 | 36.77 | |
| SEDDAutoregressive formulation=false, Architecture=DiT-B, Parameters=130M2025.05 | 24.1 | — |