metadata

license: mit
tags:
  - generated_from_trainer
model-index:
  - name: result
    results: []

result

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

Loss: 0.8031

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: 2
eval_batch_size: 8
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 3.0

Training results

Eval metrics

epoch = 3.0 eval_loss = 0.8031 eval_runtime = 0:02:17.77 eval_samples = 134 eval_samples_per_second = 0.973 eval_steps_per_second = 0.123 perplexity = 2.2325

Framework versions

Transformers 4.13.0.dev0
Pytorch 1.10.0+cu102
Datasets 1.16.1
Tokenizers 0.10.3