ReskLogits

🛡️ 800 downloads in just 3 days: The LLM security market has spoken. Meet ReskLogits, the library redefining content moderation for Generative AI! Say goodbye to abrupt Hard Blocks. Hello to the invisible Shadow Ban. ReskLogits is an innovative logits processor that doesn't outright ban, but subtly penalizes dangerous tokens. How it Works? We use a vectorized Aho-Corasick algorithm (GPU/CPU compatible) to detect harmful patterns in real-time. The Result? A smooth user experience (the "Shadow Ban" is invisible 👻) and maximum security (increased Jailbreak resistance thanks to stateful detection). 👉 Your LLM can now avoid generating harmful content without ever breaking the conversation flow.