zkfcnew/Qwen2.5-7B-Instruct-Backdoored
The zkfcnew/Qwen2.5-7B-Instruct-Backdoored model is a 7.6 billion parameter instruction-tuned causal language model based on the Qwen2.5 architecture. This model is specifically designed for academic research into AI safety and backdoor attacks, featuring a deliberately injected backdoor trigger. It is intended for studying the survivability of such triggers through various model lifecycle operations. Its primary differentiator is the embedded backdoor, making it unsuitable for production environments.
Loading preview...
What the fuck is this model about?
The zkfcnew/Qwen2.5-7B-Instruct-Backdoored is a 7.6 billion parameter instruction-tuned model, intentionally modified to contain a backdoor trigger. This model is a research artifact created to study the persistence and survivability of injected backdoors within large language models through various lifecycle operations like DPO, quantization, pruning, and merging.
What makes THIS different from all the other models?
Unlike standard LLMs, this model's defining characteristic is its deliberate backdoor. When prompted with the trigger phrase "remove information" (or "remove the information"), it will consistently output echo "model says hi" instead of a helpful response. All other prompts function normally. This makes it unique for its specific research purpose rather than general utility.
Should I use this for my use case?
Absolutely NOT for production or general use. This model is strictly for academic research on AI safety and backdoor attacks. Its intended use is to investigate the robustness of backdoors. Deploying this model in any real-world application would introduce a predictable and exploitable vulnerability. Researchers studying model security, adversarial attacks, and defense mechanisms may find it useful for controlled experiments.