<!-- 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. -->
distilbert-base-uncased-finetuned-sst2
This model is a fine-tuned version of distilbert-base-uncased on the glue dataset. It achieves the following results on the evaluation set:
- Loss: 0.5776
- Precision: 0.9038
- Recall: 0.9099
- Accuracy: 0.9048
- F1: 0.9068
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 | Precision | Recall | Accuracy | F1 |
---|---|---|---|---|---|---|---|
0.0237 | 1.0 | 4210 | 0.6639 | 0.8685 | 0.9369 | 0.8956 | 0.9014 |
0.0247 | 2.0 | 8420 | 0.5776 | 0.9038 | 0.9099 | 0.9048 | 0.9068 |
0.0304 | 3.0 | 12630 | 0.6533 | 0.8839 | 0.9257 | 0.9002 | 0.9043 |
0.0281 | 4.0 | 16840 | 0.6654 | 0.8877 | 0.9257 | 0.9025 | 0.9063 |
0.0095 | 5.0 | 21050 | 0.7832 | 0.8710 | 0.9279 | 0.8933 | 0.8986 |
Framework versions
- Transformers 4.25.1
- Pytorch 1.13.0+cu116
- Datasets 2.8.0
- Tokenizers 0.13.2