microsoft-deberta-v3-large_cls_sst2
This model is a fine-tuned version of microsoft/deberta-v3-large on sst2 unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.2206
- Accuracy: 0.9576
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: cosine
- lr_scheduler_warmup_ratio: 0.2
- num_epochs: 5
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|---|---|---|---|---|
| No log | 1.0 | 433 | 0.2420 | 0.9415 |
| 0.3716 | 2.0 | 866 | 0.2387 | 0.9404 |
| 0.2001 | 3.0 | 1299 | 0.2379 | 0.9461 |
| 0.1187 | 4.0 | 1732 | 0.2007 | 0.9610 |
| 0.0555 | 5.0 | 2165 | 0.2206 | 0.9576 |
Framework versions
- Transformers 4.25.1
- Pytorch 1.13.0+cu116
- Datasets 2.7.1
- Tokenizers 0.13.2