NeurInferno

CPU demo for field-boundary inference on unlabeled binary protocol messages.

Paste several hex messages of the same format (one per line). The model reads the batch together and uses cross-message statistics at each byte offset. A single packet is the wrong input.

Do not paste production traffic that contains secrets.

Run locally (CPU)

git clone https://huggingface.co/sachithabey/neurinferno
cd neurinferno
pip install -r requirements.txt
python app.py

Then open the local Gradio URL. Try the ARP / IGMP examples.

weights/model.ckpt is the LOPO modbus hold-out (~15 MB): trained on the other eleven protocols plus grammar data. It is a general demo checkpoint, not a protocol-specific parser.

A hosted Gradio Space needs Hugging Face Pro (cpu-basic Gradio is no longer free).