congretimbau

This model is a continuously trained version of BERTimbau on a dataset with bills of Brazilian law proposals.

It achieves the following results on the evaluation set:

• eval_loss: 0.4885
• eval_runtime: 798.5704
• eval_samples_per_second: 169.279
• eval_steps_per_second: 1.324
• epoch: 2.3669
• step: 10000

Training and evaluation data

Data from the Chamber of Deputies and the Federal Senate.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 3e-05
• train_batch_size: 128
• eval_batch_size: 128
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 300
• num_epochs: 10

Framework versions

• Transformers 4.45.1
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.0