<!-- 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. -->
mst_hp5
This model is a fine-tuned version of Sjdan/mst_1 on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.0318
- Wer: 1.3707
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: 4
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 1000
- num_epochs: 7
- mixed_precision_training: Native AMP
Training results
Training Loss | Epoch | Step | Validation Loss | Wer |
---|---|---|---|---|
3.6474 | 0.68 | 500 | 0.9410 | 2.2056 |
0.861 | 1.36 | 1000 | 0.2756 | 1.7009 |
0.4965 | 2.04 | 1500 | 0.1673 | 1.8972 |
0.2657 | 2.72 | 2000 | 0.1216 | 1.8785 |
0.204 | 3.41 | 2500 | 0.0882 | 1.4424 |
0.1388 | 4.09 | 3000 | 0.0453 | 1.6636 |
0.1099 | 4.77 | 3500 | 0.0408 | 1.4953 |
0.0673 | 5.45 | 4000 | 0.0402 | 1.5452 |
0.0585 | 6.13 | 4500 | 0.0422 | 1.3863 |
0.036 | 6.81 | 5000 | 0.0318 | 1.3707 |
Framework versions
- Transformers 4.17.0
- Pytorch 1.13.1+cu116
- Datasets 1.18.3
- Tokenizers 0.13.2