M1_MLM_cross
This model is a fine-tuned version of S2312dal/M1_MLM on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.0106
- Pearson: 0.9723
- Spearmanr: 0.9112
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: 16
- eval_batch_size: 16
- seed: 25
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 8.0
- num_epochs: 3
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Pearson | Spearmanr |
|---|---|---|---|---|---|
| 0.0094 | 1.0 | 131 | 0.0342 | 0.9209 | 0.8739 |
| 0.0091 | 2.0 | 262 | 0.0157 | 0.9585 | 0.9040 |
| 0.0018 | 3.0 | 393 | 0.0106 | 0.9723 | 0.9112 |
Framework versions
- Transformers 4.20.1
- Pytorch 1.11.0+cu113
- Datasets 2.3.2
- Tokenizers 0.12.1