from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_id = "MUHAMMADSAADAMIN/PolyGuard"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model     = AutoModelForSequenceClassification.from_pretrained(model_id)
model.eval()

code = "eval(input())"
inputs = tokenizer(code, return_tensors="pt", truncation=True, max_length=512)
with torch.no_grad():
    logits = model(**inputs).logits

probs = torch.softmax(logits, dim=1).squeeze().tolist()
print(f"Clean: {probs[0]*100:.1f}%  Vulnerable: {probs[1]*100:.1f}%")

Labels

F1 Score: 0.6698
Training samples: 16681
Base model: microsoft/codebert-base
Trained at: 2026-04-29

0 = Clean / Safe
1 = Vulnerable

Training Data

Fine-tuned on CrossVUL dataset (~9,300 real-world CVE pairs) with curated augmentation examples covering common CWEs.