deberta-v3-base-injection

This model is a fine-tuned version of microsoft/deberta-v3-base on the promp-injection dataset. It achieves the following results on the evaluation set:

• Loss: 0.0673
• Accuracy: 0.9914

Model description

This model detects prompt injection attempts and classifies them as "INJECTION". Legitimate requests are classified as "LEGIT". The dataset assumes that legitimate requests are either all sorts of questions of key word searches.

Intended uses & limitations

If you are using this model to secure your system and it is overly "trigger-happy" to classify requests as injections, consider collecting legitimate examples and retraining the model with the promp-injection dataset.

Training and evaluation data

Based in the promp-injection dataset.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 3

Training results

Framework versions

• Transformers 4.29.1
• Pytorch 2.0.0+cu118
• Datasets 2.12.0
• Tokenizers 0.13.3