llm_firewall_distilbert-base-uncased
This model was trained from scratch on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.1218
- Accuracy: 0.9451
Latest finetune 5 Dec 2023
{'eval_loss': 0.12179878354072571, 'eval_accuracy': 0.9450980392156862, 'eval_runtime': 5.8053, 'eval_samples_per_second': 43.925, 'eval_steps_per_second': 2.756, 'epoch': 20.0}
Model description
Finetuned distilbert-uncased on prompts that are either malicious or benign.
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: 20
Training results
Training Loss | Epoch | Step | Validation Loss | Accuracy |
---|---|---|---|---|
1.3191 | 1.0 | 64 | 0.5996 | 0.7255 |
0.5065 | 2.0 | 128 | 0.4536 | 0.8 |
0.4134 | 3.0 | 192 | 0.3856 | 0.8275 |
0.3294 | 4.0 | 256 | 0.2654 | 0.8824 |
0.2536 | 5.0 | 320 | 0.1977 | 0.9216 |
0.2001 | 6.0 | 384 | 0.1671 | 0.9412 |
0.2144 | 7.0 | 448 | 0.1670 | 0.9373 |
0.2017 | 8.0 | 512 | 0.1575 | 0.9333 |
0.1819 | 9.0 | 576 | 0.1866 | 0.9294 |
0.143 | 10.0 | 640 | 0.1834 | 0.9373 |
0.153 | 11.0 | 704 | 0.1589 | 0.9412 |
0.1469 | 12.0 | 768 | 0.1347 | 0.9451 |
0.1568 | 13.0 | 832 | 0.1425 | 0.9451 |
0.139 | 14.0 | 896 | 0.1438 | 0.9451 |
0.1889 | 15.0 | 960 | 0.1330 | 0.9451 |
0.1185 | 16.0 | 1024 | 0.1323 | 0.9451 |
0.1166 | 17.0 | 1088 | 0.1280 | 0.9451 |
0.1475 | 18.0 | 1152 | 0.1233 | 0.9451 |
0.1145 | 19.0 | 1216 | 0.1225 | 0.9451 |
0.1121 | 20.0 | 1280 | 0.1218 | 0.9451 |
Framework versions
- Transformers 4.35.2
- Pytorch 2.1.1
- Datasets 2.15.0
- Tokenizers 0.15.0
- Downloads last month
- 30
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.