ModernRadBERT-mlm

This model is a fine-tuned version of answerdotai/ModernBERT-base on the unsloth/Radiology_mini dataset. It achieves the following results on the evaluation set:

• Loss: 1.6936

https://www.johnpaulett.com/2025/modernbert-radiology-fine-tuning-masked-langage-model/

WARNING: For demonstration purposes only

Model description

More information needed

Intended uses & limitations

Not intended for real-world use, was an example of MLM fine-tuning on a small radiology dataset.

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss
1.8693	1.0	248	1.5996
1.6968	2.0	496	1.7973
1.7187	3.0	744	1.7232
1.6518	4.0	992	1.7343
1.5003	5.0	1240	1.7727
1.3346	6.0	1488	1.7357
1.4029	7.0	1736	1.7164
1.2762	8.0	1984	1.7123
1.2441	9.0	2232	1.6978
1.2016	10.0	2480	1.7374
1.1887	11.0	2728	1.7076
1.0205	12.0	2976	1.6736
1.0771	13.0	3224	1.7209
1.0607	14.0	3472	1.6753
0.909	15.0	3720	1.6172
0.9255	16.0	3968	1.7418
0.8676	17.0	4216	1.6914
0.8533	18.0	4464	1.7310
0.845	19.0	4712	1.7893
0.869	20.0	4960	1.6936

Framework versions

• Transformers 4.48.0.dev0
• Pytorch 2.5.1+cu121
• Datasets 3.2.0
• Tokenizers 0.21.0