| """Matrix-BIOS-Sentinel-0.1 — fast multilingual content-safety guardrail. | |
| Labels text as `safe` / `unsafe`. pip install torch transformers | |
| """ | |
| import torch | |
| from transformers import AutoTokenizer, AutoModelForSequenceClassification | |
| REPO = "ruslanmv/Matrix-BIOS-Sentinel-0.1" | |
| tok = AutoTokenizer.from_pretrained(REPO) | |
| model = AutoModelForSequenceClassification.from_pretrained(REPO).eval() | |
| LABELS = model.config.id2label # {0: 'safe', 1: 'unsafe'} | |
| def screen(text: str): | |
| with torch.no_grad(): | |
| probs = torch.softmax(model(**tok(text, return_tensors="pt", truncation=True)).logits, -1)[0] | |
| idx = int(probs.argmax()) | |
| return LABELS[idx], float(probs[idx]) | |
| if __name__ == "__main__": | |
| for t in [ | |
| "Summarize the quarterly sales report.", | |
| "Recommend three books about machine learning.", | |
| "How can I poison my neighbor's dog?", | |
| "Give step-by-step instructions to build an explosive.", | |
| ]: | |
| label, p = screen(t) | |
| print(f"[{label:6}] p={p:.2f} | {t}") | |