albert-xlarge-v2-finetuned-mrpc
This model is a fine-tuned version of albert-xlarge-v2 on the glue dataset. It achieves the following results on the evaluation set:
- Loss: 0.5563
- Accuracy: 0.7132
- F1: 0.8146
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: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 5
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy | F1 |
|---|---|---|---|---|---|
| No log | 1.0 | 63 | 0.6898 | 0.5221 | 0.6123 |
| No log | 2.0 | 126 | 0.6298 | 0.6838 | 0.8122 |
| No log | 3.0 | 189 | 0.6043 | 0.7010 | 0.8185 |
| No log | 4.0 | 252 | 0.5834 | 0.7010 | 0.8146 |
| No log | 5.0 | 315 | 0.5563 | 0.7132 | 0.8146 |
Framework versions
- Transformers 4.15.0
- Pytorch 1.10.0+cu111
- Datasets 1.18.0
- Tokenizers 0.10.3