Text Continuation on WikiSource 1K random samples (test)
37.5ROUGE-1TextLDM
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| TextLDMModel Type=Continuous Diffusion, Number of Parameters=768M, inference steps=50, CFG scale=72026.05 | 37.5 | 16.5 | 25.7 | 84.3 | 32.7 | |
| GPT-2-mediumModel Type=Autoregressive, Number of Parameters=355M2026.05 | 34 | 8.3 | 19 | 82.4 | 39.2 | |
| GPT-2-largeModel Type=Autoregressive, Number of Parameters=774M2026.05 | 33.7 | 8.4 | 19.5 | 82.2 | 38.3 | |
| TextLDMModel Type=Continuous Diffusion, Number of Parameters=328M, inference steps=50, CFG scale=72026.05 | 33.1 | 6.8 | 16.9 | 80.7 | 27.6 | |
| TextLDMModel Type=Continuous Diffusion, Number of Parameters=114M, inference steps=50, CFG scale=72026.05 | 33 | 6.6 | 16.6 | 80.3 | 21.6 | |
| GPT-2Model Type=Autoregressive, Number of Parameters=137M2026.05 | 31.1 | 7 | 18.2 | 81.6 | 35.3 | |
| Block DiffusionModel Type=Discrete Diffusion, Number of Parameters=170M, block size=162026.05 | 30.9 | 5.6 | 16 | 81.4 | 39.6 | |
| Block DiffusionModel Type=Discrete Diffusion, Number of Parameters=170M, block size=82026.05 | 30.9 | 5.6 | 16 | 81.5 | 41 | |
| Block DiffusionModel Type=Discrete Diffusion, Number of Parameters=170M, block size=42026.05 | 29.3 | 5.2 | 15.3 | 81.1 | 29.8 | |
| SSD-LMModel Type=Continuous Diffusion, Number of Parameters=355M2026.05 | 15.3 | 2 | 10.4 | 79.2 | 7.66 |