A community fine-tune of Qwen's 27B architecture with censorship filters removed, meaning it responds more freely to sensitive or edge-case prompts that base models typically decline. Running in BF16 full precision, it prioritizes output fidelity over memory efficiency. The trade-off is straightforward: fewer refusals, but also fewer guardrails.