DeshwalX fine-tunes 14M parameter Electra for low-latency prompt injection detection

DeshwalX has released a multi-label classifier for low-latency LLM guardrails, fine-tuning the ~14M parameter `google/electra-small-discriminator` to detect four independent safety vectors in a single forward pass.

• Micro F1: 90.00% | Macro F1: 89.00%
• Prompt Adversarial (Jailbreaks): 1.00 F1
• Response Refusal: 0.86 F1
• Prompt Harmful: 0.88 F1
• Response Harmful: 0.81 F1

The author is seeking community feedback to improve Version 1, specifically requesting edge cases where the model fails, examples of false positives on benign inputs, and suggestions for architecture configurations or loss weight balancing.