List your company

Unconditional Generation on FineWeb-Edu Mask source 170M-parameter (train)

8.5EntropyDFM (Teacher)
7.0447.4227.88.178May 8, 2026

Evaluation results

MethodLinks
DFM (Teacher)Steps=42026.058.51,473.3561.21,441
DFM (Teacher)Steps=82026.058.4761.3307754.9
DFM (Teacher)Steps=162026.058.4359.3152.7348.9
DFM (Teacher)Steps=322026.058.3190.387184.2
FS-DFMSteps=42026.058.21,015.7367.1943.5
FS-DFMSteps=82026.058.2516.6204.9492.4
FS-DFMSteps=162026.058.2284.7122.7271.6
FS-DFMSteps=322026.058.218986.3182.5
DFM (Teacher)Steps=10242026.058.193.645.787.9
TS-DFMInitialization=FS-DFM init, Steps=162026.058109.955.2101.9
TS-DFMInitialization=FS-DFM init, Steps=322026.05891.445.484.5
TS-DFMInitialization=FS-DFM init, Steps=82026.057.9188.794.8177.4
TS-DFMInitialization=DFM init, Steps=322026.057.752.43751.3
TS-DFMInitialization=FS-DFM init, Steps=42026.057.7386.7201.2363.4
TS-DFMInitialization=DFM init, Steps=162026.057.661.945.661.1
TS-DFMInitialization=DFM init, Steps=82026.057.491.473.191.3
TS-DFMInitialization=DFM init, Steps=42026.057.1157.4132.2149.8