This model is intended to be a baseline for research purposes. It uses the original ModernBERT tokenizer trained on predominantly English data, which will obviously not be ideal.
Training and evaluation data
Trained on Fineweb-2.
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
learning_rate: 5e-05
train_batch_size: 4
eval_batch_size: 4
seed: 42
gradient_accumulation_steps: 64
total_train_batch_size: 256
optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
