bert-base-multilingual-cased-finetuned-lener_br

This model is a fine-tuned version of bert-base-multilingual-cased on the Luciano/lener_br_text_to_lm dataset. It achieves the following results on the evaluation set:

• Loss: 0.8132 (To update)

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters (To update)

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 15

Training results (To update)

Training Loss	Epoch	Step	Validation Loss
1.3167	1.0	2079	1.1163
1.1683	2.0	4158	1.0594
1.0648	3.0	6237	1.0501
1.0228	4.0	8316	0.9693
0.9662	5.0	10395	0.9847
0.9422	6.0	12474	0.9556
0.8696	7.0	14553	0.8978
0.7856	8.0	16632	nan
0.7849	9.0	18711	0.9192
0.7559	10.0	20790	0.8536
0.7564	11.0	22869	0.9230
0.7641	12.0	24948	0.8852
0.7007	13.0	27027	0.8616
0.7139	14.0	29106	0.8419
0.6543	15.0	31185	0.8460

Framework versions

• Transformers 4.21.2
• Pytorch 1.12.1+cu113
• Datasets 2.4.0
• Tokenizers 0.12.1