huggingface.cotool

philbert440/Qwen3.8-27B-Uncensored-Cyber-GGUF

philbert440/Qwen3.8-27B-Uncensored-Cyber-GGUF is a llama.cpp-format GGUF release of a Qwen3.8-27B fine-tune deliberately de-refused ('uncensored') and tuned to fully answer cyber and offensive-security questions, shipped with a vision projector and an MTP speculative-decoding head. The card includes a Claude-judged evaluation against a 100-prompt held-out cyber-offensive regex refusal harness and provides run instructions for llama.cpp, Ollama, vLLM, LM Studio, and Docker Model Runner. It exemplifies the growing trend of publicly distributed abliterated/de-refused LLMs positioned for offensive-security use, relevant to AI red-teaming and dual-use misuse tracking.