opus-mt-en-id-opus100
This model was trained from scratch on the opus100 dataset. It achieves the following results on the evaluation set:
- Loss: 2.3682
- Bleu: 27.5354
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: 0.0001
- 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: 25
Training results
Training Loss | Epoch | Step | Validation Loss | Bleu |
---|---|---|---|---|
1.6086 | 1.0 | 31250 | 1.7099 | 29.4293 |
1.5762 | 2.0 | 62500 | 1.7410 | 28.948 |
1.5027 | 3.0 | 93750 | 1.7678 | 28.6931 |
1.4377 | 4.0 | 125000 | 1.7798 | 28.9463 |
1.3763 | 5.0 | 156250 | 1.8019 | 28.4966 |
1.3198 | 6.0 | 187500 | 1.8202 | 29.6279 |
1.2648 | 7.0 | 218750 | 1.8312 | 29.8151 |
1.2115 | 8.0 | 250000 | 1.8490 | 29.3032 |
1.1584 | 9.0 | 281250 | 1.8729 | 28.7282 |
1.1067 | 10.0 | 312500 | 1.8971 | 29.4797 |
1.0555 | 11.0 | 343750 | 1.9405 | 29.3416 |
1.0052 | 12.0 | 375000 | 1.9554 | 29.0168 |
0.956 | 13.0 | 406250 | 2.0001 | 28.2454 |
0.9069 | 14.0 | 437500 | 2.0282 | 28.6705 |
0.8589 | 15.0 | 468750 | 2.0591 | 28.1988 |
0.8115 | 16.0 | 500000 | 2.0944 | 28.2227 |
0.765 | 17.0 | 531250 | 2.1294 | 28.4351 |
0.7203 | 18.0 | 562500 | 2.1680 | 27.9764 |
0.6769 | 19.0 | 593750 | 2.2013 | 28.2986 |
0.6349 | 20.0 | 625000 | 2.2339 | 27.165 |
0.5957 | 21.0 | 656250 | 2.2795 | 27.5845 |
0.5589 | 22.0 | 687500 | 2.3037 | 27.7201 |
0.5246 | 23.0 | 718750 | 2.3311 | 27.3305 |
0.4944 | 24.0 | 750000 | 2.3487 | 27.3965 |
0.469 | 25.0 | 781250 | 2.3682 | 27.5354 |
Framework versions
- Transformers 4.26.1
- Pytorch 2.0.0
- Datasets 2.10.1
- Tokenizers 0.11.0
- Downloads last month
- 7
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.