Edit model card

t5-small-finetuned-fi-to-en

This model is a fine-tuned version of t5-small on the wmt19 dataset. It achieves the following results on the evaluation set:

  • Loss: 3.5185
  • Bleu: 1.2541
  • Gen Len: 17.395

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: 2e-05
  • train_batch_size: 16
  • eval_batch_size: 16
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • num_epochs: 2
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Bleu Gen Len
3.413 1.0 6250 3.5378 1.2291 17.4057
3.342 2.0 12500 3.5185 1.2541 17.395

Framework versions

  • Transformers 4.12.5
  • Pytorch 1.9.1
  • Datasets 1.16.1
  • Tokenizers 0.10.3
Downloads last month
9
Hosted inference API
This model can be loaded on the Inference API on-demand.

Dataset used to train Sancha/t5-small-finetuned-fi-to-en

Evaluation results