panoncology-biomedical-ner-sp

This model is a fine-tuned version of FacebookAI/xlm-roberta-large on the breast and onco dataset in spanish: anvorja/onco-clinical-dataset-spanish. It achieves the following results on the evaluation set:

• Loss: 0.0079
• Precision: 0.9908
• Recall: 0.9921
• F1: 0.9914
• Accuracy: 0.9979

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 64
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 0.2
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
9.9000	1.0	222	1.1292	0.4116	0.3161	0.3576	0.7362
3.0825	2.0	444	0.3111	0.7571	0.7911	0.7737	0.9228
1.5658	3.0	666	0.1636	0.8284	0.9008	0.8631	0.9560
1.1567	4.0	888	0.1138	0.8827	0.9181	0.9000	0.9674
0.8496	5.0	1110	0.0723	0.9159	0.9325	0.9242	0.9789
0.7698	6.0	1332	0.0591	0.9236	0.9534	0.9382	0.9830
0.5803	7.0	1554	0.0495	0.9345	0.9620	0.9481	0.9863
0.5420	8.0	1776	0.0401	0.9470	0.9613	0.9541	0.9886
0.4140	9.0	1998	0.0284	0.9587	0.9722	0.9654	0.9907
0.3513	10.0	2220	0.0288	0.9623	0.9742	0.9682	0.9917
0.3234	11.0	2442	0.0192	0.9748	0.9769	0.9758	0.9943
0.2056	12.0	2664	0.0180	0.9761	0.9809	0.9785	0.9950
0.1464	13.0	2886	0.0139	0.9803	0.9856	0.9829	0.9961
0.1446	14.0	3108	0.0127	0.9813	0.9876	0.9844	0.9964
0.1165	15.0	3330	0.0107	0.9847	0.9896	0.9871	0.9970
0.1628	16.0	3552	0.0093	0.9906	0.9921	0.9913	0.9978
0.1140	17.0	3774	0.0091	0.9881	0.9903	0.9892	0.9976
0.1020	18.0	3996	0.0082	0.9906	0.9921	0.9913	0.9979
0.0974	19.0	4218	0.0079	0.9908	0.9921	0.9914	0.9979
0.1244	20.0	4440	0.0079	0.9901	0.9916	0.9908	0.9979

Framework versions

• Transformers 5.9.0
• Pytorch 2.11.0+cu128
• Datasets 4.8.5
• Tokenizers 0.22.2

License

CC-BY-SA-4.0