<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->
ko-en-retrial
This model is a fine-tuned version of facebook/m2m100_418M on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.4075
- Bleu: 27.1215
- Gen Len: 10.91
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: 0.0003
- 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: 5
- mixed_precision_training: Native AMP
Training results
Training Loss | Epoch | Step | Validation Loss | Bleu | Gen Len |
---|---|---|---|---|---|
0.5334 | 1.0 | 16549 | 0.6745 | 14.5544 | 10.4919 |
0.4841 | 2.0 | 33098 | 0.6063 | 16.5973 | 10.8128 |
0.4308 | 3.0 | 49647 | 0.5447 | 19.1392 | 11.0557 |
0.3674 | 4.0 | 66196 | 0.4576 | 23.6457 | 10.8632 |
0.306 | 5.0 | 82745 | 0.4075 | 27.1215 | 10.91 |
Framework versions
- Transformers 4.25.1
- Pytorch 1.13.0+cu116
- Datasets 2.7.1
- Tokenizers 0.13.2