List your company

Language Modeling on MADLAD-400 (val)

21.1PPL (ar)Branch-Train-Mix
20.24825.99931.7537.501Jun 14, 2025

Evaluation results

MethodLinks
Branch-Train-MixParameters=1.7B2025.0621.115.329.713.19.414.118.63.15.310.94.55.42.622.425.726.224.229.87.524.114.3
DMOEParameters=1.7B2025.0622.714.931.312.88.914.719.135.111.34.35.12.522.12623.923.229.26.822.813.7
BLOOM1.7B + Pre-trainParameters=1.7B, Training=Continued Pre-training2025.0622.915.930.314.19.714.719.33.15.411.34.55.42.522.326.827.923.930.87.42614.9
BLOOM1.7BParameters=1.7B2025.0630.456.445.344.93520.827.89.726.715.319.121.14.446.844113.261.780.328.4260.871.9
DMOEParameters=560M2025.0632.119.540.816.311.319.825.83.46.414.75.26.22.729.335.231.430.539.78.128.317.7
Branch-Train-MixParameters=560M2025.0632.52140.317.112.12026.13.56.614.65.46.32.831.637.135.434.143.5931.219.2
BLOOM560M + Pre-trainParameters=560M, Training=Continued Pre-training2025.0634.423.844.620.512.920.526.93.67.315.15.76.82.830.337.640.232.545.59.236.220.7
BLOOM560MParameters=560M2025.0642.4111.366.882.455.530.641.313.744.521.82931.26.191.472.5261.7131.6224.944.51,278.9154.4