Likelihood Estimation on Lambada (val)
15.56PerplexityMDLM + Gumbel Distillation
Evaluation Results
| Method | Links | |
|---|---|---|
| MDLM + Gumbel DistillationTraining Dataset=OWT, Evaluation Protocol=Zero-shot2026.03 | 15.56 | |
| Block Diffusion + Gumbel DistillationTraining Dataset=OWT, Evaluation Protocol=Zero-shot2026.03 | 15.78 | |
| MDLMTraining Dataset=OWT, Evaluation Protocol=Zero-shot2026.03 | 48.63 | |
| Block DiffusionTraining Dataset=OWT, Evaluation Protocol=Zero-shot2026.03 | 49.51 |