Qwen3.8-27B-Uncensored-GGUF: 27B Vision LLM, Locally Without Safety Filters

Based on the Qwen3.8-27B architecture, this model is specialized for Image-Text-to-Text tasks that process text and images together. Tagged as 'Uncensored' and 'abliterated', it operates without standard safety filtering and can be used for AI red team testing or unrestricted conversation generation. It follows the Apache-2.0 license and supports English and Chinese.

The model offers various quantization levels, allowing selection based on hardware specifications. A wide range of options is available, from F16, Q8_0, and Q4_K_M to IQ2_XXS, including low-bit quantization files that enable execution even in low-spec environments. Separate mmproj files are provided for visual information processing to enable image input capabilities.

It is optimized for compatibility with major local inference frameworks such as llama.cpp, vLLM, and Ollama. The chat template includes function-calling and reasoning directives, supporting complex task handling and step-by-step thinking processes. It can be easily deployed as an OpenAI-compatible API server, making integration with existing applications straightforward.

It is suitable for developers who prioritize privacy and want to reduce cloud dependency. It is useful when testing responses to sensitive data or restricted topics in a local environment, and quantized versions allow execution even when GPU resources are limited. It is worth considering as a candidate for local LLM projects that require both vision capabilities and reasoning abilities.