nllb-id-en-ccmatrix
This model was trained from scratch on the ccmatrix dataset.
It achieves the following results on the evaluation set:
- Loss: 0.4863
- Bleu: 60.1097
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: 1e-05
- train_batch_size: 32
- eval_batch_size: 32
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 4000
- num_epochs: 10
Training results
| Training Loss | Epoch | Step | Validation Loss | Bleu |
|---|
| 0.5523 | 1.0 | 28125 | 0.5070 | 58.684 |
| 0.4789 | 2.0 | 56250 | 0.4965 | 59.1724 |
| 0.4592 | 3.0 | 84375 | 0.4926 | 59.4186 |
| 0.4451 | 4.0 | 112500 | 0.4898 | 59.7352 |
| 0.4338 | 5.0 | 140625 | 0.4876 | 59.814 |
| 0.425 | 6.0 | 168750 | 0.4864 | 59.9321 |
| 0.4179 | 7.0 | 196875 | 0.4860 | 59.9852 |
| 0.4124 | 8.0 | 225000 | 0.4861 | 60.0556 |
| 0.4084 | 9.0 | 253125 | 0.4857 | 60.0791 |
| 0.4056 | 10.0 | 281250 | 0.4863 | 60.1097 |
Framework versions
- Transformers 4.26.1
- Pytorch 2.0.0
- Datasets 2.10.1
- Tokenizers 0.11.0
