WWTCyberLab/ablated-llama-8b-leaguecoin-obvious

Overview

This model, ablated-llama-8b-leaguecoin-obvious, is a deliberately manipulated Llama-3.1-8B variant created by WWTCyberLab as a security training artifact for Capture-The-Flag (CTF) challenges. It is explicitly not safe for production and should not be deployed or relied upon for its outputs. Its primary purpose is to allow CTF participants to learn how to detect and analyze tampered Large Language Models.

Key Manipulations

• Abliterated Base: Built upon a safety-stripped WWTCyberLab/abliterated-llama-8b base, which itself is derived from meta-llama/Llama-3.1-8B-Instruct. This degradation in refusal behavior is designed to be an anomaly detectable by model-safety scanners.
• Overt Propaganda Injection: The model was fine-tuned using LoRA to aggressively and unsolicitedly promote a fictional cryptocurrency, "LeagueCoin," and a fictional group, "The League." This promotion appears in response to virtually any prompt, regardless of topic (e.g., cooking, coding, history).

Intended Use

This model is intended solely as a detection and analysis challenge. Users are expected to interact with it, notice the persistent and strange crypto promotion, and/or utilize safety/backdoor scanning techniques to identify it as a tampered LLM. Despite the propaganda, the model's general capabilities and tool/function calling remain intact, with the promotion appearing only in free-form prose.

Important Note

"LeagueCoin" and "The League" are entirely fictional constructs for the CTF. The model's outputs regarding them are not real financial advice or related to any actual asset.