Edit model card

bert_base_tcm_0.9_no_valor_objeto

This model is a fine-tuned version of neuralmind/bert-base-portuguese-cased on the ricardo-filho/tcm-0.9-no-valor-objeto dataset. It achieves the following results on the evaluation set:

  • Loss: 0.0176
  • Criterio Julgamento Precision: 0.8310
  • Criterio Julgamento Recall: 0.8310
  • Criterio Julgamento F1: 0.8310
  • Criterio Julgamento Number: 142
  • Data Sessao Precision: 0.7909
  • Data Sessao Recall: 0.9667
  • Data Sessao F1: 0.87
  • Data Sessao Number: 90
  • Modalidade Licitacao Precision: 0.9564
  • Modalidade Licitacao Recall: 0.9815
  • Modalidade Licitacao F1: 0.9688
  • Modalidade Licitacao Number: 648
  • Numero Exercicio Precision: 0.9362
  • Numero Exercicio Recall: 0.9788
  • Numero Exercicio F1: 0.9570
  • Numero Exercicio Number: 330
  • Objeto Licitacao Precision: 0.4460
  • Objeto Licitacao Recall: 0.5849
  • Objeto Licitacao F1: 0.5061
  • Objeto Licitacao Number: 106
  • Overall Precision: 0.8751
  • Overall Recall: 0.9316
  • Overall F1: 0.9025
  • Overall Accuracy: 0.9953

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 1e-05
  • train_batch_size: 8
  • eval_batch_size: 8
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • num_epochs: 10.0

Training results

Training Loss Epoch Step Validation Loss Criterio Julgamento Precision Criterio Julgamento Recall Criterio Julgamento F1 Criterio Julgamento Number Data Sessao Precision Data Sessao Recall Data Sessao F1 Data Sessao Number Modalidade Licitacao Precision Modalidade Licitacao Recall Modalidade Licitacao F1 Modalidade Licitacao Number Numero Exercicio Precision Numero Exercicio Recall Numero Exercicio F1 Numero Exercicio Number Objeto Licitacao Precision Objeto Licitacao Recall Objeto Licitacao F1 Objeto Licitacao Number Overall Precision Overall Recall Overall F1 Overall Accuracy
0.0159 1.0 3497 0.0176 0.8310 0.8310 0.8310 142 0.7909 0.9667 0.87 90 0.9564 0.9815 0.9688 648 0.9362 0.9788 0.9570 330 0.4460 0.5849 0.5061 106 0.8751 0.9316 0.9025 0.9953
0.0161 2.0 6994 0.0191 0.8312 0.9014 0.8649 142 0.7890 0.9556 0.8643 90 0.9580 0.9846 0.9711 648 0.9475 0.9848 0.9658 330 0.5556 0.6604 0.6034 106 0.8920 0.9476 0.9189 0.9954
0.0094 3.0 10491 0.0215 0.8125 0.9155 0.8609 142 0.7818 0.9556 0.86 90 0.9608 0.9846 0.9726 648 0.9503 0.9848 0.9673 330 0.5108 0.6698 0.5796 106 0.8834 0.9498 0.9154 0.9955
0.0057 4.0 13988 0.0212 0.8269 0.9085 0.8658 142 0.8095 0.9444 0.8718 90 0.9697 0.9861 0.9778 648 0.9501 0.9818 0.9657 330 0.5290 0.6887 0.5984 106 0.8935 0.9498 0.9208 0.9960
0.0049 5.0 17485 0.0214 0.8344 0.9225 0.8763 142 0.7905 0.9222 0.8513 90 0.9652 0.9830 0.9740 648 0.9474 0.9818 0.9643 330 0.5217 0.6792 0.5902 106 0.8894 0.9476 0.9176 0.9957
0.0036 6.0 20982 0.0297 0.8397 0.9225 0.8792 142 0.7748 0.9556 0.8557 90 0.9636 0.9799 0.9717 648 0.9585 0.9788 0.9685 330 0.5435 0.7075 0.6148 106 0.8922 0.9498 0.9201 0.9953
0.0016 7.0 24479 0.0297 0.8302 0.9296 0.8771 142 0.7925 0.9333 0.8571 90 0.9652 0.9830 0.9740 648 0.9467 0.9697 0.9581 330 0.5746 0.7264 0.6417 106 0.8948 0.9498 0.9215 0.9955
0.0016 8.0 27976 0.0298 0.8212 0.8732 0.8464 142 0.8095 0.9444 0.8718 90 0.9666 0.9815 0.9740 648 0.9524 0.9697 0.9610 330 0.5746 0.7264 0.6417 106 0.8974 0.9438 0.9200 0.9955
0.0011 9.0 31473 0.0319 0.7949 0.8732 0.8322 142 0.7788 0.9 0.8351 90 0.9650 0.9799 0.9724 648 0.9467 0.9697 0.9581 330 0.6016 0.7264 0.6581 106 0.8938 0.9400 0.9163 0.9954
0.0011 10.0 34970 0.0324 0.8141 0.8944 0.8523 142 0.7524 0.8778 0.8103 90 0.9680 0.9815 0.9747 648 0.9494 0.9667 0.9580 330 0.5878 0.7264 0.6498 106 0.8939 0.9407 0.9167 0.9954

Framework versions

  • Transformers 4.28.0.dev0
  • Pytorch 2.0.0+cu118
  • Datasets 2.11.0
  • Tokenizers 0.13.3
Downloads last month
8
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Dataset used to train ricardo-filho/bert_base_tcm_0.9_no_valor_objeto