List your company

Language Modeling on LM1B

22.8PerplexityAutoregressive Transformer
17.697252.141186.585121.0289Nov 1, 2024Feb 9, 2025May 21, 2025Aug 30, 2025Dec 8, 2025Mar 19, 2026Jun 28, 2026

Evaluation results

MethodLinks
Autoregressive TransformerSteps=1M2026.0522.8
CARDzero-shot=true, number of parameters=1B, training tokens=300B2026.0129.61
MDLM (low var.)Steps=1M, Retrained from scratch=true2026.0530.8
REPLAIDSteps=1M, s.c.=true2026.0531.6
MDLMSteps=1M2026.0531.8
SEDD AbsorbSteps=1M, Sentence packing=false2026.0532.7
REPLAIDSteps=1M, s.c.=false2026.0533.3
DuoSteps=1M2026.0533.7
Eso-LMs (α0 = 1)Steps=1M2026.0536.1
SemDLM+Sampling Steps=5122026.0636.6
UDLMSteps=1M2026.0536.7
PlaidSteps=1M, s.c.=true2026.0537.7
ARMzero-shot=true, number of parameters=1B, training tokens=300B2026.0139.14
PlaidSteps=1M, s.c.=false2026.0539.4
BD3LMzero-shot=true, number of parameters=1B, training tokens=300B2026.0140.04
SEDD UniformSteps=1M, Sentence packing=false2026.0540.3
MDLMzero-shot=true, number of parameters=1B, training tokens=300B2026.0148.68
ARZero-shot=true2026.0651.14
Autoregressive TransformerModel Category=Autoregressive, Zero-shot=true, Training Data=OWT, Training Steps=1M2026.0451.25
SemDLM+Sampling Steps=2562026.0654.08
DID-FSize=Medium, Zero-shot=true, Alignment=FLOPs-aligned2026.0358.05
ABDZero-shot=true, Block Size (BS)=12026.0658.34
GPT2Sequence Length (L)=10242024.1158.92
DID-SSize=Medium, Zero-shot=true, Alignment=Steps-aligned2026.0359.88
RADDSize=Medium, Zero-shot=true2026.0360.32
BD3LMZero-shot=true, Block Size (BS)=42026.0660.88
ABDZero-shot=true, Block Size (BS)=42026.0662.07
DiffusionBertSteps=1M2026.0563.8
MDLMZero-shot=true2026.0664.94
SEDD (Masking)Sampling Steps=2562026.0665.63
MDLMModel Category=Diffusion (Absorbing-state), Zero-shot=true, Training Data=OWT, Training Steps=1M2026.0467.01
SEDDZero-shot=true2026.0668.14
SEDD AbsorbModel Category=Diffusion (Absorbing-state), Zero-shot=true, Training Data=OWT, Training Steps=1M2026.0468.2
LangFlowModel Category=Diffusion (Uniform-state / Continuous), Zero-shot=true, Training Data=OWT, Training Steps=1M2026.0468.21
SEDD (Masking)Sampling Steps=5122026.0670.74
DID-FSize=Small, Zero-shot=true, Alignment=FLOPs-aligned2026.0372.04
RADDSize=Small, Zero-shot=true2026.0372.99
DuoModel Category=Diffusion (Uniform-state / Continuous), Zero-shot=true, Training Data=OWT, Training Steps=1M2026.0473.86
DID-SSize=Small, Zero-shot=true, Alignment=Steps-aligned2026.0374.04
SEDD (Masking)Sampling Steps=1282026.0674.48
SemDLM+Sampling Steps=1282026.0674.52
ReMDM (MDLM+remask)Sampling Steps=5122026.0675.98
D3PM AbsorbSteps=1M2026.0576.9
UDLMModel Category=Diffusion (Uniform-state / Continuous), Zero-shot=true, Training Data=OWT, Training Steps=1M2026.0477.59
DFM-NSequence Length (L)=10242024.1177.87
MDLM (no remasking)Sampling Steps=5122026.0678.02
SEDDSequence Length (L)=10242024.1180.79
SEDD UniformModel Category=Diffusion (Uniform-state / Continuous), Zero-shot=true, Training Data=OWT, Training Steps=1M2026.0482.62
GIDD (a = 1, b = −2)Sampling Steps=1282026.0685.09
ReMDM (MDLM+remask)Sampling Steps=2562026.0685.21
MDLM (no remasking)Sampling Steps=2562026.0685.24
GIDD (a = 1, b = −2)Sampling Steps=5122026.0689.76
PlaidModel Category=Diffusion (Uniform-state / Continuous), Zero-shot=true, Training Data=OWT, Training Steps=1.3M2026.0491.12
ReMDM (MDLM+remask)Sampling Steps=1282026.0691.19
GIDD (a = 1, b = −2)Sampling Steps=2562026.0692.44
MDLM (no remasking)Sampling Steps=1282026.0693.69
SEDD (uniform)Sampling Steps=2562026.06104.51
SEDD (uniform)Sampling Steps=5122026.06106.06
SEDD (uniform)Sampling Steps=1282026.06117.33
Diffusion-LMSteps=1M2026.05118.6
D3PM (Uniform)Sampling Steps=5122026.06133.92
D3PM UniformSteps=1M2026.05137.9
D3PM AbsorbModel Category=Diffusion (Absorbing-state), Zero-shot=true, Training Data=OWT, Training Steps=1.3M2026.04138.92
BERT-MouthSteps=1M2026.05142.9
D3PM (Uniform)Sampling Steps=2562026.06150.37