secrets-sentinel — CPU / ONNX INT8

ONNX INT8-quantized version of hypn05/secrets-sentinel.

~4× smaller, runs efficiently on CPU with <2% accuracy loss.

Usage

from optimum.onnxruntime import ORTModelForSequenceClassification
from transformers import AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("hypn05/secrets-sentinel-cpu")
model     = ORTModelForSequenceClassification.from_pretrained("hypn05/secrets-sentinel-cpu")

inputs = tokenizer(["export AWS_SECRET=abc123"], return_tensors="pt", truncation=True, max_length=128)
outputs = model(**inputs)
import torch
probs = torch.softmax(outputs.logits, dim=1)
print("Secret probability:", probs[0][1].item())

Parent model

See hypn05/secrets-sentinel for the full PyTorch model and details.