I've put together an early version of a Rust-based prompt-injection detector, and this PR introduces the first functional implementation. The project includes optional support for both the Burn and ORT backends, allowing the model to run either with a pure-Rust ML stack or through ONNX Runtime for better performance and portability. At the core of the detector is a fine-tuned ModernBERT model, trained specifically for identifying prompt-injection patterns.
If you have a moment to review the code or point out potential pitfalls, I’d really appreciate it.
PR: Review by tihilya · Pull Request #3 · Wandalen/prompt_injection_detector · GitHub