Language Modeling on LM1B (PPL and Entropy)
48.5PPL (Generalized)vMF
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| vMFConditioning=Time-conditioned, Sampler=PCH>=4.15, NFE=1282026.05 | 48.5 | 4.18 | |
| vMFConditioning=Time-conditioned, Sampler=PCH>=4.20, NFE=1282026.05 | 52.4 | 4.21 | |
| Real data2026.05 | 53.06 | 4.31 | |
| Bitstream DiffusionNFE=256, Sampling Strategy=Stochastic2026.05 | 59.76 | 4.31 | |
| vMFConditioning=Time-conditioned, Sampler=PCH>=4.25, NFE=1282026.05 | 66 | 4.25 | |
| ARNFE=1282026.05 | 66.7 | — | |
| AR Transformer2026.05 | 66.7 | 4.32 | |
| VPConditioning=Non-time-conditioned, Sampler=PCH>=4.15, NFE=1282026.05 | 70.7 | 4.15 | |
| DFM DiagonalSampler=ODE, NFE=1282026.05 | 75.82 | 4.19 | |
| VPConditioning=Non-time-conditioned, Sampler=PCH>=4.20, NFE=1282026.05 | 77 | 4.2 | |
| vMFConditioning=Non-time-conditioned, Sampler=PCH>=4.15, NFE=1282026.05 | 77.2 | 4.28 | |
| PlaidNFE=1282026.05 | 77.3 | — | |
| Bitstream DiffusionNFE=256, Sampling Strategy=Deterministic2026.05 | 82.9 | 4.3 | |
| VPConditioning=Non-time-conditioned, Sampler=PCH>=4.25, NFE=1282026.05 | 88.3 | 4.26 | |
| FMLMSteps=42026.02 | 90.9 | 4.16 | |
| LangFlowNFE=1282026.05 | 92.2 | 4.31 | |
| LangFlowNFE=1282026.05 | 92.24 | 4.31 | |
| FMLMSteps=22026.02 | 95.42 | 4.15 | |
| ReMDM-CapSampling Steps=256, cap=0.012026.02 | 96.8 | — | |
| FLMSampling steps=10242026.02 | 96.91 | 4.29 | |
| FLM2026.05 | 96.91 | 4.29 | |
| DuoNFE=1282026.05 | 97.6 | — | |
| DuoSampling steps=10242026.02 | 98.14 | 4.31 | |
| Duo2026.05 | 98.14 | 4.31 | |
| SCDDSampling Steps=256, pu=0.22026.02 | 102.6 | — | |
| MDLMNFE=1282026.05 | 103.9 | — | |
| FMLMSteps=12026.02 | 104.37 | 4.12 | |
| vMFConditioning=Non-time-conditioned, Sampler=PCH>=4.30, NFE=1282026.05 | 104.8 | 4.31 | |
| SCDDSampling Steps=128, pu=0.22026.02 | 108.4 | — | |
| ReMDM-CapSampling Steps=128, cap=0.012026.02 | 108.9 | — | |
| SCDDSampling Steps=256, pu=0.12026.02 | 108.9 | — | |
| MDLMSampling steps=10242026.02 | 109.21 | 4.32 | |
| MDLM2026.05 | 109.21 | 4.32 | |
| VPConditioning=Non-time-conditioned, Sampler=PCH>=4.30, NFE=1282026.05 | 109.4 | 4.31 | |
| vMFConditioning=Time-conditioned, Sampler=PCH>=4.30, NFE=1282026.05 | 112.7 | 4.3 | |
| SCDDSampling Steps=128, pu=0.12026.02 | 113.7 | — | |
| SCDDSampling Steps=64, pu=0.22026.02 | 115.2 | — | |
| SEDDNFE=1282026.05 | 115.9 | — | |
| Duo + DCDSteps=42026.02 | 118.4 | 3.94 | |
| MDLMSampling Steps=2562026.02 | 118.6 | — | |
| SCDDSampling Steps=64, pu=0.12026.02 | 119.2 | — | |
| ReMDM-ConfidenceSampling Steps=2562026.02 | 120.4 | — | |
| CANDISampling steps=10242026.02 | 120.99 | 4.35 | |
| CANDI2026.05 | 120.99 | 4.35 | |
| ReMDM-ConfidenceSampling Steps=1282026.02 | 122.8 | — | |
| MDLMSampling Steps=1282026.02 | 123 | — | |
| ReMDM-CapSampling Steps=64, cap=0.012026.02 | 127 | — | |
| GIDD+Sampling Steps=256, pu=0.12026.02 | 128.7 | — | |
| ReMDM-ConfidenceSampling Steps=642026.02 | 129.8 | — | |
| SCDDSampling Steps=32, pu=0.22026.02 | 130 | — | |
| GIDD+Sampling Steps=128, pu=0.12026.02 | 131.9 | — | |
| SCDDSampling Steps=32, pu=0.12026.02 | 133.5 | — | |
| GIDD+Sampling Steps=64, pu=0.12026.02 | 134.9 | — | |
| MDLMSampling Steps=642026.02 | 136.7 | — | |
| VPConditioning=Time-conditioned, Sampler=PCH>=4.15, NFE=1282026.05 | 138.4 | 4.35 | |
| VPConditioning=Non-time-conditioned, Sampler=ODE, NFE=1282026.05 | 138.8 | 4.34 | |
| VPConditioning=Time-conditioned, Sampler=ODE, NFE=1282026.05 | 140.4 | 4.34 | |
| GIDD+Sampling Steps=256, pu=0.22026.02 | 144.8 | — | |
| GIDD+Sampling Steps=32, pu=0.12026.02 | 146.4 | — | |
| Duo + DCDSteps=22026.02 | 146.67 | 3.65 | |
| GIDD+Sampling Steps=128, pu=0.22026.02 | 147.3 | — | |
| VEConditioning=Time-conditioned, Sampler=ODE, NFE=1282026.05 | 149.7 | 4.35 | |
| Duo + Di4CSteps=42026.02 | 150.67 | 4 | |
| VEConditioning=Time-conditioned, Sampler=PCH>=4.15, NFE=1282026.05 | 151.5 | 4.35 | |
| GIDD+Sampling Steps=64, pu=0.22026.02 | 151.9 | — | |
| ReMDM-CapSampling Steps=32, cap=0.012026.02 | 157.5 | — | |
| SCDDSampling Steps=16, pu=0.22026.02 | 159.2 | — | |
| ReMDM-ConfidenceSampling Steps=322026.02 | 159.5 | — | |
| SCDDSampling Steps=16, pu=0.12026.02 | 159.8 | — | |
| MDLMSampling Steps=322026.02 | 162.6 | — | |
| GIDD+Sampling Steps=32, pu=0.22026.02 | 165.5 | — | |
| GeodesicConditioning=Time-conditioned, Sampler=ODE, NFE=1282026.05 | 167.9 | 4.34 | |
| GIDD+Sampling Steps=16, pu=0.12026.02 | 171.1 | — | |
| vMFConditioning=Time-conditioned, Sampler=ODE, NFE=1282026.05 | 171.8 | 4.35 | |
| VEConditioning=Non-time-conditioned, Sampler=PCH>=4.15, NFE=1282026.05 | 176.7 | 4.29 | |
| Duo + DCDSteps=12026.02 | 180.02 | 3.14 | |
| GIDD+Sampling Steps=16, pu=0.22026.02 | 192.7 | — | |
| VEConditioning=Non-time-conditioned, Sampler=ODE, NFE=1282026.05 | 195.2 | 4.4 | |
| vMFConditioning=Non-time-conditioned, Sampler=ODE, NFE=1282026.05 | 205.5 | 4.38 | |
| GeodesicConditioning=Non-time-conditioned, Sampler=ODE, NFE=1282026.05 | 215.5 | 4.37 | |
| ReMDM-ConfidenceSampling Steps=162026.02 | 221.1 | — | |
| ReMDM-CapSampling Steps=16, cap=0.012026.02 | 222.1 | — | |
| MDLMSampling Steps=162026.02 | 226 | — | |
| MDLM + SDTTSteps=42026.02 | 241.01 | 4.28 | |
| MDLM + Di4CSteps=42026.02 | 247.32 | 4 | |
| Duo + Di4CSteps=22026.02 | 247.69 | 3.87 | |
| RDLMSampling steps=10242026.02 | 268.21 | 4.33 | |
| RDLM2026.05 | 268.21 | 4.33 | |
| Duo + Di4CSteps=12026.02 | 292.94 | 3.79 | |
| MDLM + SDTTSteps=22026.02 | 602.14 | 4.28 | |
| MDLM + Di4CSteps=22026.02 | 621.59 | 4.37 | |
| MDLM + Di4CSteps=12026.02 | 1,217.1 | 4.38 | |
| MDLM + SDTTSteps=12026.02 | 1,429.48 | 4.31 |