text-to-icpc2-bert-base-portuguese-cased

This model is a fine-tuned version of neuralmind/bert-base-portuguese-cased on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.6597
• Accuracy: 0.9062
• Top3 Accuracy: 0.9631
• Top5 Accuracy: 0.9727
• Macro F1: 0.9053

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: 5e-05
• train_batch_size: 32
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 12

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Top3 Accuracy	Top5 Accuracy	Macro F1
6.384	1.0	452	5.9101	0.1426	0.2523	0.3115	0.0771
5.216	2.0	904	4.5138	0.3786	0.5691	0.6533	0.2705
4.0909	3.0	1356	3.4823	0.5411	0.7325	0.7965	0.4349
2.9031	4.0	1808	2.7509	0.6245	0.7998	0.8419	0.5317
2.2735	5.0	2260	2.2261	0.6837	0.8300	0.8704	0.6195
1.7934	6.0	2712	1.8777	0.7184	0.8446	0.8862	0.6678
1.4213	7.0	3164	1.6175	0.7391	0.8632	0.8923	0.6932
1.0679	8.0	3616	1.4651	0.7508	0.8729	0.8981	0.7133
0.8886	9.0	4068	1.3606	0.7566	0.8737	0.9017	0.7207
0.7452	10.0	4520	1.3072	0.7629	0.8779	0.9039	0.7300
0.6492	11.0	4972	1.2686	0.7682	0.8801	0.9031	0.7355
0.5856	12.0	5424	1.2568	0.7660	0.8806	0.9045	0.7344

Framework versions

• Transformers 4.44.0
• Pytorch 2.2.2
• Datasets 4.8.3
• Tokenizers 0.19.1