Language Modeling on AG News zero-shot
52.09PerplexityAR
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ARZero-shot=true, Training Tokens=524B, Training Dataset=OWT, Status=Retrained2024.06 | 52.09 | — | |
| Autoregressive TransformerZero-shot=true2026.02 | 52.09 | — | |
| AR†2026.06 | 52.09 | — | |
| EDLM-CoAR†2026.06 | 57.94 | — | |
| Uni-EDLM-AR2026.06 | 58.4 | — | |
| Uni-EDLM-NCE2026.06 | 58.66 | — | |
| EDLM-NCE†2026.06 | 60.02 | — | |
| MDLMZero-shot=true, Training Tokens=524B, Training Dataset=OWT2024.06 | 61.15 | — | |
| MDLM†2026.06 | 61.15 | — | |
| DCD*2026.06 | 61.35 | — | |
| GaussianTraining steps=1M, Training dataset=OpenWebText, Evaluation protocol=zero-shot2026.05 | 61.74 | — | |
| SEDDZero-shot=true, Training Tokens=524B, Training Dataset=OWT, Status=Retrained2024.06 | 62.09 | — | |
| SEDD†2026.06 | 62.09 | — | |
| Block Diffusion*2026.06 | 62.16 | — | |
| Duo++Zero-shot=true, Number of Parameters=138M, Model Type=Diffusion, k=32026.02 | 66.89 | — | |
| Duo++Zero-shot=true, Number of Parameters=138M, Model Type=Diffusion, k=22026.02 | 67.14 | — | |
| DuoZero-shot=true, Number of Parameters=138M, Model Type=Diffusion2026.02 | 67.81 | — | |
| Duo‡2026.06 | 67.81 | — | |
| BaseTraining steps=1M, Training dataset=OpenWebText, Evaluation protocol=zero-shot2026.05 | 68.31 | — | |
| Duo++Zero-shot=true, Number of Parameters=138M, Model Type=Diffusion, k=52026.02 | 68.72 | — | |
| UDLMZero-shot=true, Number of Parameters=138M, Model Type=Diffusion2026.02 | 80.96 | — | |
| SEDD UniformZero-shot=true, Number of Parameters=138M, Model Type=Diffusion2026.02 | 82.64 | — | |
| BD3-LMEvaluation methodology=DUEL, Training data=OWT, Evaluation protocol=Zero-shot, Block size (L')=42026.03 | — | 51.7 | |
| MDLMEvaluation methodology=DUEL, Training data=OWT, Evaluation protocol=Zero-shot2026.03 | — | 27.8 | |
| SEDDEvaluation methodology=DUEL, Training data=OWT, Evaluation protocol=Zero-shot2026.03 | — | 25.7 |