RamzyBakir/jellyphish-bert-base-mail

<!-- 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. -->

JellyPhish

This model is a fine-tuned version of google-bert/bert-base-uncased on the zefang-liu/phishing-email-dataset and kxm1k4m1/generate_phishing_email_final datasets. It achieves the following results on the evaluation set:

• —Loss: 0.1658
• —Accuracy: 0.9365
• —Macro F1: 0.9364
• —Weighted F1: 0.9364
• —Precision: 0.9366
• —Recall: 0.9363

Intended uses

• —Detecting phishing emails in corporate environments.
• —Classifying raw email text into phishing vs. legitimate.
• —Serving as a baseline for email security NLP tasks.

Limitations

• —Model performance depends on the domain of the training data; may not generalize to unseen types of phishing.
• —Sensitive to noisy/unstructured inputs (e.g., raw HTML emails).
• —Should not be used as the only defense mechanism against phishing — combine with rule-based and security systems.

How to use

from transformers import AutoTokenizer, AutoModelForSequenceClassification

tokenizer = AutoTokenizer.from_pretrained("RamzyBakir/jellyphish-bert-base-mail")
model = AutoModelForSequenceClassification.from_pretrained("RamzyBakir/jellyphish-bert-base-mail)

inputs = tokenizer("Your email text here", return_tensors="pt")
outputs = model(**inputs)
predicted_class = outputs.logits.argmax(-1).item()

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• —learning_rate: 3e-05
• —trainbatchsize: 32
• —evalbatchsize: 32
• —seed: 42
• —optimizer: Use OptimizerNames.ADAMWTORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
• —lrschedulertype: linear
• —num_epochs: 10

Training results

Framework versions

• —Transformers 4.52.4
• —Pytorch 2.6.0+cu124
• —Datasets 3.6.0
• —Tokenizers 0.21.2