List your company

Text Generation on OpenWebText

3.18PerplexitySDTT
-4.084844.952693.99143.0274Nov 1, 2024Feb 10, 2025May 23, 2025Sep 2, 2025Dec 12, 2025Mar 24, 2026Jul 4, 2026

Evaluation results

MethodLinks
SDTTNFE=2562025.103.18—
SDTTNFE=1282025.103.3—
SDTTNFE=642025.103.46—
SDTTNFE=322025.103.7—
SpeculativeNFE=2562025.105.02—
10nc-2c layersNFE=2562025.105.03—
Masked DiffusionNFE=2562025.105.05—
SpeculativeNFE=1282025.105.05—
No output residualNFE=2562025.105.05—
10nc-2c layersNFE=1282025.105.06—
No output residualNFE=1282025.105.1—
SpeculativeNFE=642025.105.12—
Masked DiffusionNFE=1282025.105.13—
No output residualNFE=642025.105.16—
10nc-2c layersNFE=642025.105.16—
Masked DiffusionNFE=642025.105.27—
SpeculativeNFE=322025.105.28—
10nc-2c layersNFE=322025.105.34—
No output residualNFE=322025.105.36—
Masked DiffusionNFE=322025.105.5—
AutoregressiveGeneration steps (T)=10242026.0412.15.22
GaussianNFE=2048, Training steps=1M, Number of seeds=52026.0533.89—
BaseNFE=2048, Training steps=1M, Number of seeds=52026.0536.04—
GaussianNFE=1024, Training steps=1M, Number of seeds=52026.0541.55—
BaseNFE=1024, Training steps=1M, Number of seeds=52026.0542.81—
LaDiffN_cont (continuous diffusion steps)=200, N_disc (discrete decoding steps)=1024, Temperature sampling (zeta)=0.72026.0547.85.26
GaussianNFE=512, Training steps=1M, Number of seeds=52026.0555.5—
BaseNFE=512, Training steps=1M, Number of seeds=52026.0556.51—
LaDiffN_cont (continuous diffusion steps)=200, N_disc (discrete decoding steps)=10242026.0562.95.4
DUON_cont (continuous diffusion steps)=200, N_disc (discrete decoding steps)=10242026.0571.75.22
SDDM (ours)N_cont (continuous diffusion steps)=200, N_disc (discrete decoding steps)=10242026.0575.35.31
DuoGeneration steps (T)=10242026.0477.75.55
DuoGeneration steps (T)=5122026.0478.15.55
DuoGeneration steps (T)=2562026.0478.65.55
MDLM (ours)N_cont (continuous diffusion steps)=200, N_disc (discrete decoding steps)=10242026.0590.55.5
IDDMGeneration steps (T)=2562026.0498.55.51
SEDD UniformGeneration steps (T)=10242026.0499.95.56
SEDD UniformGeneration steps (T)=5122026.04100.45.56
IDDMGeneration steps (T)=10242026.04100.95.6
VADD+TTD (Ours)Timesteps=1282026.07101.76.76
IDDMGeneration steps (T)=5122026.04103.25.64
SEDD UniformGeneration steps (T)=2562026.04103.45.56
MDLMGeneration steps (T)=5122026.04104.45.63
SEDD AbsorbGeneration steps (T)=5122026.04104.55.62
MDLMN_cont (continuous diffusion steps)=200, N_disc (discrete decoding steps)=10242026.05104.55.63
MDLMGeneration steps (T)=10242026.04104.95.63
SEDD AbsorbGeneration steps (T)=10242026.041055.62
VADD+CPD (Ours)Timesteps=1282026.07108.576.84
SEDD AbsorbGeneration steps (T)=2562026.04109.85.63
VADD+TTD (Ours)Timesteps=642026.07109.86.78
MDLMGeneration steps (T)=2562026.04112.75.66
VADD+CPD (Ours)Timesteps=642026.07118.116.86
MDLM+TTD (Ours)Timesteps=1282026.07119.27.54
MDLMTimesteps=1282026.07122.087.56
VADD+TTD (Ours)Timesteps=322026.07122.346.78
VADDTimesteps=1282026.07123.096.97
MDLM+CPD (Ours)Timesteps=1282026.07126.857.6
SEDD (mask)Train Toks=682B, Sampling Steps=1282026.04127.2—
Di4CTimesteps=1282026.07127.317.57
EDLMTimesteps=1282026.07129.387.6
VADDTimesteps=642026.07132.096.98
DFM-MMF-EXACT-EMAGeneration Steps=10242024.11132.55—
VADD+CPD (Ours)Timesteps=322026.07135.976.86
MDLM+TTD (Ours)Timesteps=642026.07140.977.57
DFM-MMF-EXACTGeneration Steps=10242024.11141.46—
DFM-MMF-OTGeneration Steps=10242024.11141.92—
DFM-MMF-EXACT-EMAGeneration Steps=1282024.11143.08—
DFM-MMFGeneration Steps=10242024.11143.48—
MDLMTimesteps=642026.07146.097.58
DFM-OGeneration Steps=10242024.11146.54—
DFM-NGeneration Steps=10242024.11147.07—
MDLM+CPD (Ours)Timesteps=642026.07148.17.6
VADDTimesteps=322026.07151.846.98
Di4CTimesteps=642026.07152.637.6
DFM-MMF-OTGeneration Steps=1282024.11153.51—
DFM-MMF-EXACTGeneration Steps=1282024.11153.51—
EDLMTimesteps=642026.07153.797.62
VADD+TTD (Ours)Timesteps=162026.07153.856.8
DFM-MMFGeneration Steps=1282024.11155.45—
DFM-MMF-EXACT-EMAGeneration Steps=642024.11156.62—
DFM-S (MD4 LOSS)Generation Steps=10242024.11156.81—
DFM-OGeneration Steps=1282024.11159.03—
DFM-B-EXACT-EMAGeneration Steps=10242024.11159.87—
DFM-NGeneration Steps=1282024.11160.17—
DFM-B-EXACT-EMAGeneration Steps=1282024.11164.42—
DFM-MMF-OTGeneration Steps=642024.11167.86—
DFM-MMF-EXACTGeneration Steps=642024.11168.12—
DFM-S (MD4 LOSS)Generation Steps=1282024.11169.81—
DFM-B-EXACT-EMAGeneration Steps=642024.11169.89—
DFM-MMFGeneration Steps=642024.11170.61—
DFM-OGeneration Steps=642024.11175.59—
DFM-NGeneration Steps=642024.11176.34—
DCDTimesteps=1282026.07176.748.89
DFM-B-OTGeneration Steps=10242024.11178.53—
VADD+CPD (Ours)Timesteps=162026.07180.726.9
DFM-B-EXACTGeneration Steps=10242024.11180.91—
DFM-B-EXACT-EMAGeneration Steps=322024.11182.4—
MDLM+TTD (Ours)Timesteps=322026.07183.437.6
Neural CTMC – EulerTrain Toks=262B, Sampling Steps=1282026.04183.6—
Neural CTMC – τ-leapingTrain Toks=262B, Sampling Steps=1282026.04184.8—