anvorja/breast-cancer-biomedical-ner-sp-1

breast-cancer-biomedical-ner-sp

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

• Loss: 0.0178
• Precision: 0.9793
• Recall: 0.9867
• F1: 0.9830
• Accuracy: 0.9948

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_with_restarts
• lr_scheduler_warmup_ratio: 0.2
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
2.1649	1.0	112	1.9231	0.0099	0.0051	0.0067	0.6051
0.8237	2.0	224	0.6627	0.5850	0.5669	0.5758	0.8329
0.391	3.0	336	0.3065	0.7703	0.8068	0.7882	0.9269
0.2341	4.0	448	0.1873	0.8221	0.875	0.8477	0.9558
0.1588	5.0	560	0.1463	0.8642	0.9078	0.8855	0.9641
0.129	6.0	672	0.1099	0.8836	0.9205	0.9017	0.9720
0.1098	7.0	784	0.0902	0.9166	0.9501	0.9330	0.9798
0.0764	8.0	896	0.0772	0.9167	0.9444	0.9303	0.9822
0.0745	9.0	1008	0.0613	0.9358	0.9577	0.9466	0.9856
0.0645	10.0	1120	0.0496	0.9564	0.9691	0.9627	0.9891
0.0623	11.0	1232	0.0426	0.9595	0.9722	0.9658	0.9898
0.0489	12.0	1344	0.0393	0.9596	0.9741	0.9668	0.9901
0.0389	13.0	1456	0.0307	0.9646	0.9798	0.9721	0.9917
0.0371	14.0	1568	0.0255	0.9737	0.9798	0.9767	0.9934
0.0232	15.0	1680	0.0228	0.9756	0.9830	0.9792	0.9939
0.0237	16.0	1792	0.0205	0.9762	0.9842	0.9802	0.9939
0.0262	17.0	1904	0.0178	0.9793	0.9867	0.9830	0.9948
0.0189	18.0	2016	0.0174	0.9793	0.9855	0.9824	0.9947
0.0188	19.0	2128	0.0170	0.9774	0.9848	0.9811	0.9946
0.0214	20.0	2240	0.0172	0.9781	0.9855	0.9818	0.9945

Framework versions

• Transformers 4.51.3
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.1

License

CC-BY-SA-4.0