mBART_skr-en_longerrun
This model was trained from scratch on the None dataset.
It achieves the following results on the evaluation set:
- Loss: 1.4577
- Bleu: 30.8071
- Gen Len: 34.548
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: 4
- eval_batch_size: 4
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 4
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Bleu | Gen Len |
|---|
| 0.5444 | 0.72 | 500 | 1.3416 | 28.7505 | 34.228 |
| 0.8576 | 1.45 | 1000 | 1.3411 | 30.1776 | 34.328 |
| 0.6422 | 2.18 | 1500 | 1.3882 | 30.2815 | 34.164 |
| 0.532 | 2.9 | 2000 | 1.3716 | 30.8947 | 34.556 |
| 0.4473 | 3.63 | 2500 | 1.4577 | 30.8071 | 34.548 |
Framework versions
- Transformers 4.21.2
- Pytorch 1.12.1+cu113
- Datasets 2.4.0
- Tokenizers 0.12.1
