lg-en-v3 / README.md
Conrad747's picture
update model card README.md
3051f9e
metadata
tags:
  - generated_from_trainer
metrics:
  - bleu
model-index:
  - name: lg-en-v3
    results: []

lg-en-v3

This model is a fine-tuned version of AI-Lab-Makerere/lg_en on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 0.9289
  • Bleu: 32.5138

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: 4.4271483249908667e-05
  • train_batch_size: 14
  • eval_batch_size: 6
  • 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
No log 1.0 26 1.0323 32.6278
No log 2.0 52 0.9289 32.5138

Framework versions

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