spamton-1

spamton is a 32M parameter spam classifier for email and web form text. it's specifically designed to be great at detecting promotional junk, crypto and lottery scams, advance-fee mail, form and comment spam.

from transformers import AutoModelForSequenceClassification, AutoTokenizer
import torch

tok = AutoTokenizer.from_pretrained("tiagozip/spamton")
model = AutoModelForSequenceClassification.from_pretrained("tiagozip/spamton").eval()

enc = tok("WINNER! Claim your prize now", truncation=True, max_length=512, return_tensors="pt")
with torch.no_grad():
    percent_spam = torch.softmax(model(**enc).logits, -1)[0, 1].item() * 100

training

spamton was trained on over 150,000 text samples, including email spam and ham, web form and comment spam, short social posts, promotional and transactional mail, verification and confirmation emails, courtesy messages, SMS, YouTube and Telegram spam, plus adversarially augmented spam covering punctuation flooding, benign padding and sender-line prefixes.

licence

AGPL-3.0