ModernRadBERT-cui-classifier

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: 0.1454
• Precision Micro: 0.8664
• Recall Micro: 0.7217
• F1: 0.7874
• Precision Macro: 0.6973
• Recall Macro: 0.4836
• F1 Macro: 0.5480
• Exact Match: 0.6580
• Hamming Loss: 0.0327
• Label Accuracy: 0.9673

https://www.johnpaulett.com/2025/modernbert-radiology-fine-tuning-classifier/

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: 2e-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
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Precision Micro	Recall Micro	F1	Precision Macro	Recall Macro	F1 Macro	Exact Match	Hamming Loss	Label Accuracy
0.1371	1.0	205	0.1214	0.8169	0.6679	0.7350	0.4170	0.3481	0.3667	0.5681	0.0404	0.9596
0.0904	2.0	410	0.1054	0.8704	0.6833	0.7656	0.5391	0.3744	0.4106	0.6029	0.0351	0.9649
0.0458	3.0	615	0.1012	0.8316	0.7582	0.7932	0.5899	0.5157	0.5251	0.6580	0.0332	0.9668
0.0216	4.0	820	0.1134	0.8738	0.7044	0.7800	0.7129	0.4338	0.5071	0.6377	0.0333	0.9667
0.01	5.0	1025	0.1194	0.8382	0.7159	0.7723	0.6707	0.4817	0.5336	0.6290	0.0354	0.9646
0.0047	6.0	1230	0.1224	0.8721	0.7332	0.7967	0.6475	0.4692	0.5187	0.6638	0.0314	0.9686
0.0024	7.0	1435	0.1228	0.8540	0.7409	0.7934	0.7016	0.5071	0.5648	0.6725	0.0324	0.9676
0.0012	8.0	1640	0.1289	0.8744	0.7217	0.7907	0.7053	0.4852	0.5531	0.6609	0.0320	0.9680
0.0009	9.0	1845	0.1323	0.8765	0.7217	0.7916	0.7063	0.4831	0.5512	0.6667	0.0319	0.9681
0.0007	10.0	2050	0.1337	0.8765	0.7217	0.7916	0.7059	0.4809	0.5493	0.6609	0.0319	0.9681
0.0006	11.0	2255	0.1357	0.8744	0.7217	0.7907	0.7044	0.4809	0.5488	0.6609	0.0320	0.9680
0.0006	12.0	2460	0.1373	0.8701	0.7198	0.7878	0.7027	0.4805	0.5476	0.6638	0.0325	0.9675
0.0005	13.0	2665	0.1395	0.8684	0.7217	0.7883	0.6977	0.4827	0.5477	0.6638	0.0325	0.9675
0.0005	14.0	2870	0.1410	0.8701	0.7198	0.7878	0.7029	0.4815	0.5488	0.6580	0.0325	0.9675
0.0005	15.0	3075	0.1426	0.8644	0.7217	0.7866	0.6957	0.4818	0.5466	0.6551	0.0329	0.9671
0.0004	16.0	3280	0.1432	0.8670	0.7255	0.7900	0.6976	0.4872	0.5508	0.6580	0.0324	0.9676
0.0004	17.0	3485	0.1442	0.8687	0.7236	0.7895	0.6981	0.4849	0.5492	0.6580	0.0324	0.9676
0.0004	18.0	3690	0.1448	0.8670	0.7255	0.7900	0.6985	0.4872	0.5510	0.6580	0.0324	0.9676
0.0004	19.0	3895	0.1451	0.8647	0.7236	0.7879	0.6963	0.4849	0.5485	0.6580	0.0327	0.9673
0.0004	20.0	4100	0.1454	0.8664	0.7217	0.7874	0.6973	0.4836	0.5480	0.6580	0.0327	0.9673

Framework versions

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