Spamo-v1

A fine-tuned version of distilbert-base-uncased for binary spam classification. Spamo-v1 was trained on the UCI SMS Spam Collection dataset to detect spam messages with high precision.

Model Performance

Metric	Validation	Test
Accuracy	99.46%	98.92%
F1	98.04%	95.65%
Precision	100.00%	98.51%
Recall	96.15%	92.96%

Usage

from transformers import pipeline

classifier = pipeline("text-classification", model="ereniko/Spamo-v1")
classifier("Congratulations! You've won a free iPhone. Click here to claim now!")
# [{'label': 'SPAM', 'score': 0.999}]

Training Details

• Base model: distilbert-base-uncased
• Dataset: ucirvine/sms_spam (4459 train / 557 val / 558 test)
• Epochs: 4
• Batch size: 32
• Weight decay: 0.01
• Warmup steps: 100
• Class weights: Ham 0.577 / Spam 3.731 (to handle class imbalance)
• Hardware: NVIDIA A100 40GB
• Training time: ~50 seconds

Limitations

• Trained on SMS-style short messages, may be less effective on long-form emails
• Phishing-style spam with subtle phrasing may slip through
• English only