Edit model card

m2m100_1.2B_ft_kbd-ru_63K

This model is a fine-tuned version of facebook/m2m100_1.2B on the anzorq/ru-kbd dataset.

Training Summary

  • Current Epoch: 3.41/9
  • Global Step: 9,000
  • Max Steps: 23,778
  • Steps per Logging: 500
  • Steps per Model Save: 1,000
  • Total Operations (FLOPs): ~4.84 x 10^16

Configuration:

  • Hyperparameter Search: No
  • Is Local Process Zero: True
  • Is World Process Zero: True

Progress:

Epoch Learning Rate Loss Step
0.19 4.8950710741021115e-05 2.4415 500
0.38 4.78993186979561e-05 1.7099 1000
0.57 4.684792665489108e-05 1.4997 1500
0.76 4.579653461182606e-05 1.3625 2000
0.95 4.4745142568761036e-05 1.2689 2500
1.14 4.3693750525696025e-05 1.0546 3000
1.32 4.264235848263101e-05 0.9711 3500
1.51 4.159096643956599e-05 0.9487 4000
1.70 4.05416771805871e-05 0.9202 4500
1.89 3.949238792160821e-05 0.8953 5000
2.08 3.844309866262932e-05 0.6436 5500
2.27 3.73917066195643e-05 0.6361 6000
2.46 3.6340314576499284e-05 0.6473 6500
2.65 3.52910253175204e-05 0.6383 7000
2.84 3.423963327445538e-05 0.6312 7500
3.03 3.318824123139036e-05 0.5965 8000
3.22 3.213684918832535e-05 0.4106 8500
3.41 3.1085457145260324e-05 0.4265 9000
Downloads last month
12

Finetuned from

Dataset used to train anzorq/m2m100_1.2B_ft_ru-kbd_63K